Inertial measurement unit-based degenerative brain disease prevention system and GNSS-IMU-based wandering detection system

Abstract

The present invention relates to a wearable device-based early prediction system for collecting movement data of a user in real time by using an inertial measurement unit and analyzing the movement data through a neural network-based analysis module, thereby detecting an early sign of a degenerative brain disease. When the present invention is used, excellent gait data suitable for prevention of degenerative brain disease can be collected using a low-cost inertial measurement unit. Alternatively, the present invention relates to a system and a method for collecting movement path and gait data from a user to which both a GNSS-based positioning device and an IMU-based inertial measurement unit are attached, and analyzing the movement path and the gait data in an integrated manner through neural network analysis to determine whether wandering occurs in real time.

Description

Inertial Measurement Unit-based Degenerative Brain Disease Prevention System and GNSS-IMU-based Wandering Detection System

[0001] The present invention relates to a system for preventing the progression of degenerative brain disease using a low-cost inertial measurement device and a method for providing information on the prevention of degenerative brain disease.

[0002] Alternatively, the present invention relates to a loitering detection system and a loitering detection method that determine in real time whether a user is loitering by analyzing the user's movement path and walking characteristics using a satellite-based positioning device (GNSS) and an inertial measurement unit (IMU).

[0003]

[0004] Neurodegenerative diseases are chronic and progressive conditions caused by the degeneration of the nervous system, with Parkinson's disease and Alzheimer's disease being representative examples. These diseases lead to a decline in the ability to perform daily activities due to the deterioration and loss of nerve cell function, and in severe cases, render even autonomous physical activity impossible. As global population aging accelerates, the incidence of neurodegenerative diseases is rising rapidly, leading to increased social and economic burdens. Consequently, practical and continuous management to slow disease progression and prevent onset is becoming increasingly important.

[0005] Regular physical exercise and management through appropriate feedback are crucial for the prevention of neurodegenerative diseases. Exercise contributes not only to maintaining physical flexibility, balance, and muscle strength but also to promoting neuroplasticity in the brain. In particular, gait training, balance exercises, and aerobic activity are effective in slowing the decline in motor function—a major symptom of neurodegenerative diseases—and preventing the worsening of symptoms. However, personalized exercise prescriptions and real-time feedback are essential for effectively performing these exercises and managing them consistently.

[0006] Conventional exercise prescription and management systems are limited to providing general exercise programs and have limitations in offering personalized feedback based on real-time analysis of individual physical conditions and exercise capabilities. Furthermore, patients find it difficult to maintain proper posture and exercise methods on their own, and without expert guidance, there is a risk that symptoms may worsen due to improper exercise performance. Since these conventional systems are primarily available only in hospitals or specialized rehabilitation centers, accessibility has been low and continuous management has been difficult.

[0007] In particular, conventional gait and motion analysis systems require expensive motion capture equipment, optical sensor-based systems, or precise pressure sensor devices. While these devices offer high accuracy, they have disadvantages such as high installation and maintenance costs, as well as the need for a large space and skilled personnel for operation. Consequently, there are significant practical limitations to using them in ordinary homes or small-scale rehabilitation facilities, making it difficult for patients to utilize them consistently in their daily lives.

[0008] Furthermore, existing healthcare and body correction devices focus on collecting and analyzing data from specific body parts. As a result, they fail to provide comprehensive data, such as overall body movement or neural signals, and struggle with continuous data analysis to detect early signs of disease. Due to these issues, there is currently a lack of an integrated, real-time approach for personalized disease prediction and prevention.

[0009] Accordingly, the inventors have completed the present invention by devising a system that combines a low-cost inertial measurement unit (IMU) with artificial intelligence (AI)-based neural network analysis technology to replace existing expensive equipment while analyzing user movements in real time and providing customized exercises and feedback for the prevention of degenerative brain diseases.

[0010] Degenerative brain diseases cause cognitive impairments such as memory loss, loss of spatial perception, and weakened sense of direction, leading to frequent involuntary wandering. In particular, conditions like Alzheimer's disease and Parkinson's dementia pose a high risk of accidents or disappearances due to getting lost even in familiar spaces or moving aimlessly, requiring continuous monitoring by caregivers and nursing facilities. In response to this social demand, the importance of wandering detection technology, which monitors patient movement in real time and detects abnormal movements, is increasing significantly.

[0011] Conventional wandering detection technologies have primarily utilized GNSS-based location tracking devices or ZigBee and RFID-based access detection devices. However, GNSS-only methods face the problem of satellite signals weakening or being interrupted in indoor or urban environments, making it difficult to reliably determine the patient's actual movement path. Additionally, there is a limitation in that false alarms occur when normal movement is mistaken for wandering due to momentary noise and deviations in location signals.

[0012] Meanwhile, there are conventional technologies that utilize an Inertial Measurement Unit (IMU), but IMU data has a problem in that the trajectory gradually becomes inconsistent with the actual path during long-term movement due to drift issues. In addition, IMU-only analysis has technical limitations in that it cannot accurately reflect movement patterns unique to wandering, such as disorientation and deviation from the destination.

[0013] Furthermore, existing systems relied on methods that simply combined GNSS and IMU data or analyzed only changes in distance and speed, resulting in accuracy limitations such as failing to adequately reflect individual users' usual movement habits, mistaking instantaneous movement deviations for loitering, or conversely failing to detect dangerous movements. Additionally, there was a problem where data quality deteriorated significantly even with slight changes in the sensor's mounting position or orientation; however, conventional technology failed to provide sufficient structural functions to correct for such data consistency and reliability.

[0014] As such, wandering in patients with degenerative brain diseases is a behavioral characteristic involving a combination of complex and non-linear factors, such as individual gait characteristics, changes in movement speed, and deviation patterns from destinations; therefore, accurate judgment is difficult based solely on simple location tracking or speed analysis. Consequently, there is a need for technology capable of distinguishing between intentional movement and unintentional wandering with high accuracy by integrating and analyzing satellite-based location data and inertial sensor-based gait data using neural networks, and by generating reference patterns through the long-term accumulation and learning of user-specific movement patterns.

[0015] To meet these technical requirements, the present invention aims to provide a new type of loitering detection system that integrates neural network analysis technology fusing GNSS and IMU data, user-specific reference movement pattern learning, loitering judgment algorithm, data reliability correction function, and dead reckoning-based path estimation and map matching correction technology when satellite signals are disconnected.

[0016]

[0017] The technical problem to be solved by the present invention is to provide an inertial measurement device-based degenerative brain disease prevention system characterized by comprising: a data collection module that measures acceleration and angular velocity data of an inertial measurement device attached to a user's body part in real time and collects the measured data; a neural network analysis module that calculates gait data by analyzing the data collected through the data collection module using a neural network analysis model; a gait analysis module that detects the degree of progression of degenerative brain disease or abnormal gait characteristics by analyzing the gait data; and a feedback provision module that provides real-time feedback when abnormal gait characteristics are detected.

[0018] Another technical objective of the present invention is to provide a method for providing information to optimally control an inertial measurement device-based early prediction system for degenerative brain diseases.

[0019] The technical problem to be solved by the present invention is to provide a location measurement device and inertial measurement device-based wandering detection system characterized by comprising: a data collection module that collects location information data obtained through a satellite-based position measurement device (GNSS) attached to a user's body part and acceleration and angular velocity data obtained through an inertial measurement device (IMU) in real time; a neural network analysis module including a movement path calculation unit that calculates movement path data by analyzing the location information data through a neural network and a walking data calculation unit that calculates walking data by analyzing the acceleration and angular velocity data through a neural network; a walking analysis module that generates cognitive decline data indicating the degree of progression of a degenerative brain disease by analyzing the walking data; a wandering determination module including a wandering determination unit that determines whether the user's movement corresponds to wandering by analyzing the movement path data and the cognitive decline data; and a feedback provision module that provides one or more of an alarm, route guidance, and return guidance information in real time to the user and guardian terminal when wandering is determined by the wandering determination module.

[0020] Another technical objective of the present invention is to provide a loitering detection method that provides information for optimally performing the operation of the loitering detection system.

[0021] Other technical problems that the present invention aims to solve are not limited to those mentioned above, and other unmentioned technical problems will be clearly understood by those skilled in the art to which the present invention belongs from the description below.

[0022]

[0023] As one embodiment for achieving the above technical objective, the present invention provides an inertial measurement device-based degenerative brain disease prevention system characterized by comprising: a data collection module that measures acceleration and angular velocity data of an inertial measurement device attached to a user's body part in real time and collects the measured data; a neural network analysis module that calculates gait data by analyzing the data collected through the data collection module using a neural network analysis model; a gait analysis module that detects the degree of progression of a degenerative brain disease or abnormal gait characteristics by analyzing the gait data; and a feedback provision module that provides real-time feedback when abnormal gait characteristics are detected.

[0024] The above system may additionally include a content provision module that provides customized exercise content tailored to the user's condition.

[0025] In another aspect, the present invention provides a method for providing information to optimally control an inertial measurement device-based early prevention system for degenerative brain diseases.

[0026] First, the data collection module of the present invention can collect user movement data in real time using an Inertial Measurement Unit (IMU) sensor. The inertial measurement unit sensor may be attached in multiple suitable locations among one or more body parts selected from the group consisting of the user's shoulders, back, knees or / and feet, and more preferably, may be attached to specific locations on the spine (C7~T1), the superior angle of the scapula, the rib cage, and the knees or / and feet. Through this, the 3-axis acceleration and angular velocity data of the inertial measurement unit can be accurately measured.

[0027] The collected data is grouped into data collection time intervals between 0.001 seconds and 1 hour, processed in the form of a data window, and can be transmitted to a neural network analysis module. This data provides biomechanical information of the user, such as joint angles and movement speeds, and can be used as basic data for analyzing movement patterns.

[0028] In addition, to maintain data continuity and consistency, the data collection module can organize the measured data into a consistent format through time synchronization and transmit it to the neural network analysis module.

[0029] The data collection module described above can continuously record data of customized exercises performed by the user through the content provision module. When performing specific gait training or balance maintenance exercises, the data collection module can measure changes in acceleration and angular velocity occurring during each exercise movement. The measured data is processed by a neural network analysis module and can be converted into one or more detailed exercise data selected from a group consisting of walking speed, joint amplitude range, step count, gait cycle, presence of foot dragging, essential tremor, stride length, cadence, bilateral lower limb contact, joint range of motion, neck joint range of motion, back joint range of motion, and knee joint range of motion.

[0030] The neural network analysis module of the present invention can produce walking data by analyzing data collected through the data collection module using a neural network analysis model.

[0031] Specifically, the neural network analysis module processes data received from the data collection module using deep learning-based analysis technology, and the analysis module may include a reliability correction unit that corrects the reliability of gait data through a neural network model trained based on existing standard data measured from a plurality of conventional inertial measurement devices and a motion capture system. The reliability correction unit can improve the consistency and accuracy of the data by preprocessing the data collected from the inertial measurement devices and performing data interpolation, axis correction, outlier removal, feature scaling, and normalization.

[0032] In addition, the present module analyzes data by adopting a CNN model with a data window size of 2 to 1000, and more preferably, can analyze with a CNN model with a data window size of 40 to 60. The most preferred data window size of the present invention may be set to 50, but is not limited thereto.

[0033] Additionally, the neural network analysis module may include a position correction unit that analyzes changes in reliability and validity according to the attachment location of the inertial measurement device, performs repeatability and concurrent validity evaluations with existing standard data to select a location with high reliability and validity, and guides the user to the selected attachment location through the feedback provision module.

[0034] The above position correction unit can provide a guide to the user in real time to adjust the sensor attachment position through the feedback providing module when the reliability and validity of the inertial measurement device drop below a threshold. The threshold may be 0.6 to 0.8 or less, and more preferably set to 0.7, but is not limited thereto.

[0035] The neural network analysis module described above can analyze data collected from an inertial measurement unit to calculate one or more gait data selected from a group including walking speed, joint amplitude range, step count, gait cycle, presence of foot dragging, essential tremor, stride length, cadence, bilateral lower limb contact, joint range of motion, neck joint range of motion, back joint range of motion, and knee joint range of motion. Specifically, walking speed, gait cycle, and bilateral lower limb contact can be identified as key indicators of the gait data. When analyzing by identifying walking speed, gait cycle, and bilateral lower limb contact as key indicators of the gait data, there may be a technical advantage in that it is possible to rapidly detect early motor function decline in degenerative brain disease and efficiently provide feedback accordingly. Walking speed, gait cycle, and bilateral lower limb contact are closely related to major gait characteristics that show changes first in patients with degenerative brain disease, and by identifying and analyzing these indicators, data processing time can be shortened and analysis accuracy can be improved.

[0036] The aforementioned gait data can play an important role in detecting early signs of disease by precisely analyzing the user's movement patterns. In particular, the neural network analysis module performs a process of removing outliers and standardizing the data to increase data reliability, thereby improving the reliability and accuracy of the analysis results.

[0037] The above gait analysis module can analyze the gait data to evaluate the degree of progression of degenerative brain disease and detect abnormal gait characteristics.

[0038] The above feedback providing module may provide feedback to the user in one or more ways selected from the group consisting of vibration, visual notification, and auditory notification according to abnormal signs in the gait data calculated by the gait analysis module, but is not limited thereto.

[0039] The above feedback analyzes the gait data and provides feedback when it is above or above a specified threshold, or below or below a specified threshold, and can correct and / or improve posture and / or gait through one or more selected from the group consisting of posture correction and gait correction, but is not limited thereto.

[0040] The above feedback providing module continuously monitors the user's walking data and can progressively optimize personalized feedback based on walking patterns and changes analyzed over time.

[0041]

[0042] As one embodiment for achieving the above technical objective, the present invention provides an inertial measurement device-based degenerative brain disease prevention system characterized by comprising: a content providing module that provides customized content tailored to the user's condition; a data collection module that measures acceleration and angular velocity data of an inertial measurement device attached to a part of the user's body in real time and collects the measured data; a neural network analysis module that analyzes the data collected through the data collection module using a neural network analysis model to produce gait data and exercise data; and a gait analysis module that analyzes the gait data and exercise data.

[0043] The technical configuration and details for implementing the above-mentioned inertial measurement device-based degenerative brain disease prevention system may be applied identically or similarly to the inertial measurement device-based degenerative brain disease prevention system including the content providing module that provides the customized content of the present invention, to the extent that this does not conflict with the implementation of the technical essence.

[0044] The above content provision module may include: a content management unit that classifies exercise content pre-stored on the content provision module according to one or more criteria selected from a group consisting of the type, intensity, duration, and purpose of each content, and stores and manages it in a database; a monitoring unit that continuously collects and analyzes the gait data and the exercise data, evaluates the progression status and improvement of the degenerative brain disease, and reports it to the user on a regular basis; a customized exercise generation unit that analyzes the gait data and the exercise data and provides customized exercise content suitable for the user from among the exercise content stored in the content management unit according to the analyzed results; a health program unit that provides one or more customized protocols selected from a group consisting of diet, sleep management, meditation, psychological counseling, customized nutritional supplements, walking, and running for lifestyle improvement, based on the degree of progression of the user's degenerative brain disease and one or more pieces of information selected from a group consisting of the gait data and the exercise data; and a test content provision unit that provides test content for evaluating the user's degenerative brain disease.

[0045] Customized exercise data provided through the content provision module can be systematically collected, analyzed, classified, and stored by the data collection module. The data collection module can measure acceleration and angular velocity data in real time, analyze the data in a neural network analysis module to convert it into quantified exercise data, and store it in the data collection module.

[0046]

[0047] As another embodiment for achieving the above technical objective, the present invention provides a method for providing information on the prevention of degenerative brain disease based on an inertial measurement device, characterized by comprising: a data collection step of measuring acceleration and angular velocity data of an inertial measurement device attached to a user's body part in real time and collecting said measured data; a neural network analysis step of analyzing the data collected through a data collection module using a neural network analysis model to produce gait data; a gait analysis step of analyzing said gait data to evaluate the degree of progression of degenerative brain disease and detect abnormal gait characteristics; and a feedback provision step of providing real-time feedback when said abnormal gait characteristics are detected.

[0048]

[0049] As another embodiment for achieving the above technical objective, the present invention provides a method for providing information on the prevention of degenerative brain diseases based on an inertial measurement device, characterized by comprising: a content providing step of providing customized content suitable for the user's condition; a data collection step of measuring acceleration and angular velocity data of an inertial measurement device attached to a part of the user's body in real time and collecting said measured data; a neural network analysis step of analyzing the data collected through the data collection module using a neural network analysis model to produce walking data and exercise data; and a walking analysis step of analyzing said walking data and exercise data.

[0050]

[0051] As another embodiment for achieving the above technical objective, the present invention provides an early diagnosis method for degenerative brain disease based on an inertial measurement device, characterized by comprising: a data collection step of measuring acceleration and angular velocity data in real time using an inertial measurement device attached to a user's body part and collecting the measured data; a neural network analysis step of analyzing the data collected through a data collection module using a neural network analysis model to produce gait data; a gait analysis step of analyzing the gait data to evaluate the degree of progression of degenerative brain disease and detect abnormal gait characteristics; and a feedback provision step of providing real-time feedback when abnormal gait characteristics are detected.

[0052]

[0053] The inertial measurement device-based degenerative brain disease prevention system of the present invention can be utilized to detect and prevent the progression of degenerative brain diseases at an early stage. Furthermore, by collecting and analyzing user gait data in real time, abnormal gait patterns or the degree of disease progression can be accurately evaluated, thereby providing an opportunity for early detection of the disease and taking appropriate preventive measures. In particular, the reliability and validity of the data have been enhanced by combining neural network analysis technology with an inertial measurement device, and even more accurate data can be secured through attachment position correction.

[0054] The present invention helps delay the progression of degenerative brain diseases and improve walking ability and the quality of daily life by providing users with customized exercise content and lifestyle improvement protocols. Through a feedback provision module, it provides real-time feedback that adjusts one or more selected from a group consisting of gait correction, posture correction, balance maintenance, and walking speed, thereby improving the user's gait and posture and reducing the risk of falls. In addition, exercise performance data is continuously collected and analyzed through an exercise performance data unit and a monitoring unit to regularly evaluate and report changes in the user's condition and the degree of improvement.

[0055] The present invention is designed as a wearable system that enables low-cost, non-invasive implementation capable of replacing expensive diagnostic equipment or invasive tests, and can be used continuously in daily life. This increases accessibility and reduces the economic burden, while providing highly reliable data to users and medical professionals. Furthermore, it supports comprehensive and personalized health management by providing customized exercise content and health programs through a content delivery module.

[0056] As an embodiment for achieving the above technical objective, the present invention provides a data collection module that acquires location information data through a satellite-based positioning device (GNSS) attached to a user's body part, and measures acceleration and angular velocity data through an inertial measurement unit (IMU) attached to the same body part, and collects this data in real time. In addition, the present invention may provide a neural network analysis module comprising a movement path calculation unit that calculates movement path data by analyzing the location information data via a neural network, and a walking data calculation unit that calculates walking data by analyzing the acceleration and angular velocity data via a neural network. By simultaneously securing movement path and walking information through the neural network-based analysis, the limitations of a simple location-based method are overcome, and stable data analysis becomes possible even in various environments.

[0057] As another embodiment for achieving the above technical objective, the present invention may further include a gait analysis module that analyzes the gait data to generate cognitive decline data reflecting the degree of progression of degenerative brain disease. The gait analysis module estimates the degree of cognitive decline by analyzing IMU-based features such as gait regularity, speed, stride length, and joint movement patterns via a neural network, thereby enabling the determination of wandering based on cognitive decline, which was difficult in existing technologies. Furthermore, the present invention may include a wandering determination module that determines whether a user's movement corresponds to wandering by integrating and analyzing the movement path data and the cognitive decline data, and can perform a more precise determination of wandering by calculating an integrated risk level by applying different weights to deviation indicators, decline indicators, etc.

[0058] As another embodiment for achieving the above technical objective, the present invention may further include a reference movement pattern generation unit that, in addition to the GNSS-IMU fusion-based judgment, generates a reference movement pattern for each user by accumulating and learning the movement path data and cognitive function deterioration data over time. Through this, the user's usual movement range, walking rhythm, speed pattern, etc., are learned over a long period, and if the current movement pattern differs significantly from the reference movement pattern, it can be determined as unintentional wandering. In addition, the present invention may include a path correction unit that estimates a short-term movement trajectory using an IMU-based dead reckoning method when satellite-based position signals are disconnected or their quality deteriorates, and corrects the path using a map information-based map matching method when satellite signals are re-received, thereby minimizing movement path errors even in various surrounding environments such as indoors, tunnels, and urban canyons.

[0059] In addition, the present invention may include a feedback providing module that provides one or more of an alarm, route guidance, or return guidance information in real time to a user terminal or a guardian terminal when the user's movement is determined to be wandering by the wandering determination module. Through this, appropriate guidance can be provided to the user or a dangerous situation can be immediately notified to the guardian as soon as wandering is detected, thereby significantly reducing the possibility of accidents and disappearance.

[0060] The effects of the present invention are not limited to the effects described above, and should be understood to include all effects that can be inferred from the configuration of the invention described in the detailed description of the invention or the claims.

[0061]

[0062] FIG. 1 is a schematic diagram of an inertial measurement device-based degenerative brain disease prevention system according to one embodiment of the present invention.

[0063] FIG. 2 is a schematic diagram of an inertial measurement device-based degenerative brain disease prevention system with an added content provision module according to one embodiment of the present invention.

[0064] Figure 3 is a configuration diagram showing the neural network analysis module of Figure 1 in detail.

[0065] Figure 4 is a configuration diagram showing the content provision module of Figure 2 in detail.

[0066] Figure 5 illustrates the process of applying deep learning according to an embodiment of the present invention.

[0067] FIG. 6a is a graph of the LSTM model learning result according to one embodiment of the present invention.

[0068] FIG. 6b is a graph of the CNN model training result according to one embodiment of the present invention.

[0069] FIG. 7 is an inertial measuring device according to one embodiment of the present invention.

[0070] FIG. 8 is a cradle for attaching an inertial measuring device according to one embodiment of the present invention.

[0071] FIG. 9 is a block diagram showing the overall configuration of a wandering detection system according to one embodiment of the present invention.

[0072] FIG. 10 is a block diagram showing an embodiment of a neural network analysis module according to the present invention.

[0073] FIG. 11 is a block diagram showing a configuration in which a reference movement pattern generating unit is added as an embodiment of a wandering determination module according to the present invention.

[0074] FIG. 12 is a block diagram showing a configuration in which a path correction unit is added as an embodiment of a neural network analysis module according to the present invention.

[0075]

[0076] Those skilled in the art to which the present invention pertains will understand that the invention may be implemented in modified forms without departing from the essential nature of the invention. Therefore, the disclosed embodiments should be considered in an illustrative rather than a limiting sense. The scope of the invention is defined by the claims, not by the foregoing description, and all variations within the scope of the claims should be interpreted as being included in the invention.

[0077] Furthermore, the terminology used in this specification is used to appropriately describe preferred embodiments of the present invention, and may vary depending on the intent of the user or operator, or the conventions of the field to which the present invention belongs. Accordingly, the definitions of these terms should be based on the content throughout this specification. Throughout the specification, when a part is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0078] As used herein, the term “output” may refer to a process of generating or deriving a result by processing or calculating input data. This includes the act of obtaining a specific result value through data analysis, calculation, or algorithmic processing, and is not necessarily limited to a specific form or method. The term “output” is defined in relation to the technical description of the present invention and may be implemented in various ways depending on the data processing method, analysis technique, or form of the result value. All such implementations should be interpreted as being included within the scope of the present invention.

[0079] As used herein, the term "generation" may refer to a process of creating new data, materials, structures, or results based on specific input values ​​or conditions. This may be achieved in various ways, such as physical, chemical, or data processing, and is not necessarily limited to a specific methodology or procedure. The term "generation" is defined in relation to the technical description of the present invention and may be interpreted to include all processes that may be modified in various ways depending on the purpose of implementation or environment. All such modifications and applications are considered to be included within the scope of the present invention.

[0080] As used in this specification, the term "neural network analysis model" may refer to an artificial intelligence algorithm structure that processes input data to learn specific patterns and analyzes or predicts results based thereon. This may include various neural network structures based on Artificial Neural Networks (ANNs), such as Multilayer Perceptrons (MLPs), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Long Short-Term Memory Networks (LSTMs). The term "neural network analysis model" is used to analyze the spatial and temporal characteristics of data or to learn complex relationships, and the learned model may be designed to be applicable in various environments. The neural network analysis model used in this invention includes all neural network structures and their variations that can be configured according to the purpose of data processing and analysis, and these may be interpreted as being included within the technical scope of this invention.

[0081] As used in this specification, the term "degenerative brain disease" comprehensively refers to chronic and progressive diseases resulting from the gradual degeneration of the nervous system, and may include, for example, Parkinson's disease, Alzheimer's disease, Huntington's disease, and dementia. These diseases are characterized by the gradual deterioration of an individual's motor skills, cognitive functions, or ability to perform daily activities due to the decline and loss of neuronal function. "Degenerative brain disease" may manifest in various forms depending on specific symptoms, causes, or mechanisms of onset, and in this invention, it can be used as a central concept in the process of analyzing and processing data to detect and predict early signals of these diseases. "Degenerative brain disease" encompasses major disease categories addressed in early prediction and diagnosis utilizing neural network models, and may be interpreted to include various disease types depending on the technical purpose and scope of application.

[0082] As used herein, the term "inertial measurement unit" refers to a device designed to measure a user's body movements, which may include an accelerometer and a gyroscope (angular velocity sensor) capable of measuring acceleration and angular velocity in three axes (x, y, z). An "inertial measurement unit" is primarily referred to as an IMU (Inertial Measurement Unit) and can be utilized to collect user movement data in real time to quantify physical characteristics of movement, such as position, velocity, and changes in direction. This term is not limited to the specific form or size of the sensor and can be applied in various technical implementations and environments. Additionally, it may be implemented in various modified forms, such as geomagnetic, stretch, and pressure sensors.

[0083] As used in this specification, the term "suitable" may refer to a state that conforms to a specific purpose or situation or satisfies requirements. This may be evaluated based on various criteria, such as physical characteristics, data reliability, and device performance, and is not necessarily limited to a specific format or value. In this invention, the term "suitable" is used in the context of the attachment location of the inertial measurement device, the data processing method, or the reliability of analysis results, and may refer to satisfying the conditions necessary to ensure the efficiency and accuracy of the system. Furthermore, "suitable" is a concept that may vary depending on the environment, purpose of use, or application conditions, and may be interpreted within the scope of purpose / specific realization for solving the technical problem of this invention, including all such variations and possibilities of application.

[0084] As used in this specification, the term "early prediction" may refer to a process of identifying potential events, states, or results ahead of the general time of discovery by analyzing specific data. This includes estimating or predicting future states or results based on input data and can be achieved by utilizing various technologies such as data analysis, neural network models, or machine learning algorithms. "Early prediction" focuses on enabling preventive measures by detecting the progression of a disease or the likelihood of specific events occurring in advance. The present invention is designed to detect conditions such as degenerative brain diseases early by analyzing user movement data, thereby slowing the progression of said disease and facilitating appropriate treatment or preventive measures. Furthermore, the term "early prediction" includes a process that may be modified depending on specific conditions or situations. For example, the accuracy and temporal scope of "early prediction" may vary depending on the precision of the data, the analysis algorithm, or the structure of the prediction model. All such possibilities for modification are considered to be included within the purpose / specific scope of realization for solving the technical problem of the present invention, and the early prediction function for said disease can be realized through various implementation methods.

[0085] As used herein, the term "sensing speed" refers to the cycle of measuring or collecting data through a sensor, and may mean the frequency of recording or updating data within a specific time unit. This signifies the speed at which the sensor detects and outputs information, and is generally expressed as the number of data collections per second (Hz) or the measurement interval (e.g., in seconds). "Sensing speed" affects the precision and real-time nature of data collection and may be adjusted according to the system's analysis purpose, target environment, or application field. In this specification, "sensing speed" may be included as an important technical element for ensuring the data accuracy and reliability of an early prediction system for degenerative brain diseases.

[0086] As used herein, the term "data collection time interval" may refer to the time interval during which data is collected by an inertial measurement unit (IMU), and may be specified as the time interval during which each sample is measured in a continuously collected data stream. This is a unit in which data is grouped based on a specific time period, and it can be a factor that directly affects the precision of the data and the accuracy of the analysis results.

[0087] For example, if the "data collection time interval" is set to 0.01 seconds, the IMU sensor can measure and record acceleration and angular velocity data every 0.01 seconds. By adjusting these time interval settings, detailed analysis of body movement data and reliable data processing can be achieved.

[0088] In this specification, the term "data collection time interval" may represent the same concept as "sensing rate," and both terms may equally refer to a time period for collecting data. Accordingly, "sensing rate" may be used as a synonym for "data collection time interval" in this specification and may be interpreted interchangeably depending on the context.

[0089] As used in this specification, the term "data window" may refer to a unit for processing or analyzing continuously collected data by grouping it into a specific time interval or range. This is a temporal or data block established to effectively analyze and process data, and can be used as a basic unit for organizing data collected from an inertial measurement unit (IMU) and inputting it into a model.

[0090] The above "data window" may be designed to split a continuous data stream during the data processing process or to selectively analyze data occurring in specific time intervals. In the present invention, the "data window" may represent the form of data grouped according to the sensing speed or the data collection time interval, and may be utilized as an input value to a deep learning model to improve the reliability and efficiency of analysis for the early prediction of degenerative brain diseases. Furthermore, the term "data window" may be modified in various forms depending on the specific data processing method or analysis purpose. Specifically, the method of data organization may vary depending on the window size, data grouping method, time interval setting, etc., and this may be adjusted to suit specific analysis environments or conditions. All such modifications and possibilities of application may be considered to be included within the purpose / specific scope of realization for solving the technical problem of the present invention.

[0091] As used in this specification, the term "cadence" may refer to a biomechanical parameter representing the frequency of periodic movement over a specific period of time, particularly the number of steps per minute in a user's gait cycle. This serves as key data for quantitatively evaluating a user's movement characteristics and can be utilized for gait analysis, evaluation of movement patterns, and / or early prediction of degenerative brain diseases. The term "cadence" is not necessarily limited to walking and can also be applied when measuring or evaluating repetitive movement patterns of the body. Specifically, in one embodiment of the present invention, "cadence" may be used to precisely analyze a user's walking speed, rhythm, and pattern based on data collected from an Inertial Measurement Unit (IMU). It can serve as a key analysis variable in neural network analysis modules and early prediction models, acting as an important factor in detecting signals of degenerative brain diseases or identifying gait abnormalities. Furthermore, the term "cadence" may be modified in various ways depending on the purpose of analysis or the method of data collection. This can be calculated at different frequencies or time intervals depending on specific environments or conditions, and can be used as a broad concept for measuring the frequency of repetitive movements in various physical activities. In this specification, all such applicability and variations may be considered to be included within the scope of the purpose / specific realization for solving the technical problem of the present invention.

[0092] As used in this specification, the term "reliability" may refer to the degree to which results appear consistently when data is repeatedly measured or analyzed under the same conditions. This serves as an important criterion for evaluating the reproducibility of results during the system's data collection and analysis process and can be considered essential to ensure data consistency and accuracy. In this invention, "reliability" is used to evaluate the repeatability of data collected from an Inertial Measurement Unit (IMU) and results derived during the neural network analysis process, and can serve as a key element in ensuring the stability and reliability of the system. Furthermore, "reliability" may be evaluated in various forms depending on the data collection environment, the state of the device, or the analysis conditions, and all such variations and applications may be interpreted as being included within the purpose / specific scope of realization for solving the technical problem of this invention.

[0093] As used in this specification, the term "validity" may refer to the degree to which specific data or analysis results accurately reflect the target or standard intended for actual measurement. This can be used as a criterion to evaluate not only the reliability of the data but also whether the results are substantially appropriate and meaningful. In the present invention, "validity" is evaluated based on the correlation between existing standard data (gold standard), such as Inertial Measurement Units (IMUs) and Motion Capture Systems (VICONs), and can play an important role in ensuring the validity and appropriateness of the analysis results. Furthermore, "validity" may be measured in various ways depending on the purpose of evaluation, the nature of the data, or the analysis environment, and all such variations and applicability may be considered to be included within the purpose / specific scope of realization for solving the technical problem of the present invention.

[0094] The term "CNN (Convolutional Neural Network)" as used in this specification refers to a type of deep learning that can mean a neural network structure capable of automatically extracting and learning the characteristics of input data through convolutional operations. The "CNN" is generally used for pattern recognition of image data or time-series data and has strengths in extracting data features regardless of specific location. In the present invention, the "CNN" can be utilized to learn spatial characteristics by analyzing data collected from an Inertial Measurement Unit (IMU) and thereby identify important patterns in motion data. This "CNN" model can be combined with other neural network structures to enhance the accuracy of data analysis and contribute to effectively detecting early signals of degenerative brain diseases. The term "CNN" is expandable depending on various configurations and layer combinations and can be implemented in various modified forms within the scope of the purpose / specific realization for solving the technical problem of the present invention.

[0095] As used in this specification, the term "LSTM (Long Short-Term Memory)" refers to an extended structure of a Recurrent Neural Network (RNN) that may signify a neural network model specialized for learning time-series data with strong temporal dependencies. "LSTM" is a structure designed to learn the long-term dependencies of data, utilizing cell state and gate mechanisms to appropriately reflect past information in current learning. In this specification, "LSTM" can be utilized to analyze motion data collected from an Inertial Measurement Unit (IMU) to learn temporal patterns and to detect early signals of diseases such as degenerative brain diseases. "LSTM" processes data step-by-step and can identify important temporal features in motion data by combining past and present information. The scope of application of the term "LSTM" can be expanded through various hyperparameter settings and network configurations, and it can be implemented in various forms within the purpose / specific scope of realization for solving the technical problem of the present invention.

[0096] As used herein, the term "wearable" refers to a device worn or attached to the body by a user and may include electronic or mechanical systems for collecting data, analyzing, and providing user feedback. This is a device designed for the purpose of monitoring a user's physical activity or physiological state or transmitting information, and may be implemented in various forms such as wristbands, smartwatches, and body-attached sensors. The term "wearable" is not limited to a specific form or function and may refer to a device that includes core functions such as data collection and processing and a user interface. For example, the "wearable" of the present invention may include an Inertial Measurement Unit (IMU) to perform a key role in collecting user movement data in real time and analyzing it to generate predictable information. Furthermore, wearable devices may be modified in various forms depending on the usage environment, purpose, and technical requirements. This may be a device embedded in clothing, a sensor attached to shoes, or an independent device, and all such modifications should be interpreted as being included within the scope of the present invention. "Wearables" are not merely physical devices but function as technological elements that enhance user experience and the efficiency of data utilization, and can include various implementation methods and application possibilities.

[0097] As used in this specification, the term "Python" refers to a high-level programming language that can be used to perform various tasks such as data collection, preprocessing, analysis, visualization, and training of deep learning models. It enables efficient complex data processing and analysis based on clear syntax, code readability, and powerful library support (Pandas, NumPy, TensorFlow, etc.), and facilitates system control and real-time data processing by providing platform independence and automation capabilities. In this invention, Python is utilized for neural network model development, data preprocessing, and analysis in an inertial measurement device-based degenerative brain disease prevention system, and all such variations and applications may be interpreted as being included within the scope of the purpose / specific realization for solving the technical problem of this invention.

[0098] As used in this specification, the term "abnormal gait characteristics" refers to movement characteristics that deviate from the general pattern observed during walking. These characteristics may manifest in various forms, such as reduced walking speed, limited range of motion of joints, foot dragging, irregularity of walking rhythm, and abnormal knee flexion angles, and are detected by analyzing acceleration and angular velocity data collected through an inertial measurement device. In this invention, abnormal gait characteristics are analyzed in real time, and customized feedback is provided based on this analysis. All such variations and applications may be interpreted as being included within the scope of the purpose / specific realization for solving the technical problem of this invention.

[0099] As used in this specification, the term "label" refers to a correct answer or target value corresponding to input data in machine learning and deep learning models, and can be used as a criterion for the model to learn and evaluate prediction results. In the present invention, when learning inertial measurement device data, absolute angles collected from a motion capture system are used as labels, and these are converted into a form suitable for model learning through standardization or normalization during the data preprocessing process. Labels play an important role in calculating errors and evaluating model performance by comparing them with predicted values, and all such variations and applications can be interpreted as being included within the purpose / specific scope of realization for solving the technical problem of the present invention.

[0100] As used in this specification, the term "existing standard" may refer to reference values ​​or verified data referenced to evaluate the accuracy and validity of data. In this invention, data collected from high-precision equipment, such as a motion capture system (VICON), may be used as an existing standard, but is not limited thereto. In this invention, data acquired from a motion capture system may be used as an existing standard to develop the algorithm of the reliability correction unit. The optimized algorithm data developed in this way may also become a new existing standard, and this may be utilized to verify reliability and validity in the position correction unit when a user wears an inertial measurement unit (IMU), but is not limited thereto. Furthermore, this invention performs repeatability and concurrent validity evaluations based on this existing standard data to correct the data quality of the inertial measurement unit and increase reliability. Moreover, the existing standard is used as a standard for comparison and verification during the data analysis and neural network model learning process, thereby enabling the continuous improvement of the performance of the optimized algorithm. All such variations and applications may be interpreted as being included within the purpose / specific scope of realization for solving the technical problem of this invention.

[0101] The term "RMSE (Root Mean Square Error)" as used in this specification is an error measurement metric used to quantitatively evaluate the difference between a predicted value and an actual value, and can be used to evaluate the performance of a prediction model. RMSE is a value obtained by calculating the square mean of the error and taking the square root; it intuitively expresses the magnitude of the error, and a lower value indicates higher accuracy of the prediction model. In this invention, RMSE is used to evaluate the difference between the absolute angle predicted by a neural network model based on Inertial Measurement Unit (IMU) data and existing standards (motion capture system data), thereby verifying the performance of the model and the effectiveness of data correction. Furthermore, RMSE may be modified according to various window sizes or model structures, and all such modifications and applications may be interpreted as being included within the purpose / specific scope of realization for solving the technical problem of this invention.

[0102] As used in this specification, the term "essential tremor" refers to a type of neurological disorder characterized by unintended, regular, and rhythmic tremors that primarily manifest in the hands, head, and neck. It occurs without a specific cause and is often triggered by genetic factors or abnormal activity of the nervous system. "Essential tremor" may become more pronounced during states of tension or the performance of specific movements, and can cause discomfort in daily life. In the process of analyzing and diagnosing "essential tremor," the present invention focuses on evaluating the intensity, frequency, and pattern of the tremor by utilizing user movement data and an Inertial Measurement Unit (IMU). Through this, the degree of tremor can be identified, and appropriate exercise content or vibration feedback can be provided to contribute to symptom relief. In this invention, the term "essential tremor" may be used in various clinical settings and can be interpreted in diverse contexts, such as the alleviation of tremor symptoms, early detection, and preventive management.

[0103] As used in this specification, the term "bilateral lower limb grounding" may refer to the action or state in which both lower limbs of a user simultaneously touch the ground while walking. This is an important process for ensuring stability and balance as part of the gait cycle, and it frequently occurs, particularly when walking slowly or while stationary. "Bilateral lower limb grounding" is frequently observed in patients with impaired balance or degenerative neurological diseases and can be used as an important indicator for evaluating gait stability or motor function. In this invention, to analyze the "bilateral lower limb grounding" state, information collected from gait data and an inertial measurement device can be utilized to evaluate the user's balance and gait status in real time and provide feedback. Through this, exercise content suitable for the user can be provided, or abnormal gait patterns can be detected to contribute to the early prediction and prevention of degenerative brain diseases. The term "bilateral lower limb grounding" may be used in various clinical and everyday environments and can be implemented in various ways depending on system design and application.

[0104] As used in this specification, the term "hyperparameter" refers to a variable that is set in advance by a user when training a machine learning and deep learning model, and may mean a factor that directly affects the structure of the model and the learning process. Hyperparameters play an important role in tuning and optimizing the performance of a model and are generally not learned automatically but may be determined experimentally or empirically, but are not limited thereto.

[0105] The term “walking data” as used in this specification refers to data collected through an inertial measurement device while a user walks or moves normally, and may include various indicators such as walking speed, stride length, cadence, whether the foot drags, essential tremor, and contact with both lower limbs.

[0106] Additionally, the term “exercise data” as used herein may refer to data recorded during the process of a user performing exercise content provided by a content provision module, and, similar to gait data, may include indicators such as walking speed, stride length, cadence, presence of foot dragging, essential tremor, and bilateral lower limb contact. Exercise data may be classified according to one or more criteria selected from a group consisting of the exercise body part, purpose, type, duration, intensity, and number of repetitions, and may be utilized to evaluate the user’s exercise performance status and improvement.

[0107] The aforementioned gait and exercise data interact complementarily to contribute to a comprehensive analysis of the user's condition and the provision of personalized exercise content suitable for the individual. The gait and exercise data not only reflect the user's current state but can also be utilized to design continuous feedback and protocols for the long-term prevention and improvement of degenerative brain diseases.

[0108]

[0109] Preferred embodiments according to the present invention will be described in detail below with reference to the accompanying drawings. In describing the present invention below, if it is determined that a detailed description of known functions or configurations may unnecessarily obscure the essence of the present invention, such detailed description will be omitted. Unless otherwise specifically defined, all terms in this specification have the same general meaning as understood by a person skilled in the art to which the present invention pertains. In the event of a conflict with the meaning of a term used in this specification, the definition used in this specification shall prevail.

[0110]

[0111] The inertial measurement device-based degenerative brain disease prevention system of the present invention utilizes a low-cost inertial measurement device attached to a user's body to collect body movement and gait data in real time, and based on this, can provide solutions necessary for the prevention and management of degenerative brain diseases.

[0112] FIG. 1 is a schematic diagram of an inertial measurement device-based degenerative brain disease prevention system according to one embodiment of the present invention, wherein the inertial measurement device-based degenerative brain disease prevention system may include a data collection module (100), a neural network analysis module (200), a gait analysis module (300), and a feedback provision module (400).

[0113] The data collection module (100) can perform the function of measuring and collecting acceleration and angular velocity data in real time using a low-cost inertial measurement unit (IMU) attached to the user's body. The inertial measurement unit may be configured to include a 3-axis acceleration sensor and a 3-axis angular velocity sensor. The sensors used in the present invention are not limited thereto and may be implemented as sensors of various forms and performances.

[0114] The above-described inertial measuring device may be attached to one or more body parts selected from the group consisting of the shoulder, back, knee or / and foot. Preferably, it may be attached to one or more body parts selected from the group consisting of the spine (C7–T1), superior angle of the scapula, rib cage, knee or / and foot. More preferably, it may be attached to the upper and lower parts of the knee, respectively, to precisely measure the flexion and extension angles of the knee joint during walking, but is not limited thereto.

[0115] Preferably, if the inertial measurement device is attached to the upper and lower parts of the knee, the movement trajectory and rotation angle of the knee joint during walking can be captured more accurately, and important walking characteristics such as the range of motion (ROM) can be analyzed more finely. This method is excellent for detecting inaccurate rotation or movement limitations of the knee and can play an important role in detecting early signs of degenerative brain disease. Analysis utilizing data from the upper and lower parts of the knee significantly improves the accuracy and consistency of walking data compared to the existing single-sensor method and can increase the precision of joint angle measurements. In addition, by comprehensively processing data from multiple devices, the movement speed, rotation characteristics, and trajectory changes of the joint can be evaluated more finely.

[0116] More preferably, the inertial measurement device may be attached to the upper and lower parts of the knee and simultaneously attached to one or more upper body parts selected from the group consisting of specific spinal locations (C7–T1), the superior angle of the scapula, and the rib cage. By attaching sensors to the upper body, the motor coordination and balance status between the upper and lower body can be evaluated more precisely. In particular, sensors attached to the shoulders can capture upper body sway and balance changes during walking more accurately than when sensors are attached only to the lower body. The attachment of such inertial measurement devices can enhance the reliability and validity of the degenerative brain disease prevention system by detecting movement discrepancies or abnormal patterns between the upper and lower body at an early stage.

[0117] The data collection module (100) can collect acceleration and angular velocity data by sampling at intervals of 0.001 seconds to 1 hour. Preferably, the data can be collected by sampling at intervals of 0.01 seconds, but is not limited thereto. Through such high-speed sampling, minute changes in movement can be captured, and the continuity and precision of the data can be guaranteed. The collected data is transmitted to a processing device in real time, and time synchronization is performed to increase the accuracy of the data. In addition, the collected data can be grouped at regular time intervals and processed in the form of a data window.

[0118] If the sampling interval is set to less than 0.001 seconds, an excessive amount of data is generated, which can place a significant burden on the system. As the number of data points collected per second increases rapidly, data processing and storage devices become overloaded, potentially degrading system efficiency. Furthermore, such high-speed sampling provides unnecessary precision for analyzing human movement characteristics, failing to bring substantial benefits to system performance improvement. In addition, as the amount of data increases, real-time processing in neural network analysis modules or gait analysis modules may be delayed, and noise increases due to minute sensor vibrations or electrical noise, leading to a decrease in the signal-to-noise ratio (SNR). This noise can degrade the accuracy of analysis results. Conversely, if the sampling interval exceeds one hour, subtle changes in movement may fail to be captured. Rapid movements or detailed joint motions occurring during walking are not measured, resulting in reduced data continuity and potentially requiring interpolation or correction algorithms. Such discontinuity reduces the accuracy of gait analysis or abnormal pattern detection, posing a risk of missing critical signals, particularly in cases requiring the detection of minute abnormalities such as degenerative brain diseases. In addition, if the sampling interval is prolonged, it becomes difficult to provide real-time feedback, which may result in delays in delivering immediate feedback on gait correction or balance maintenance to the user.

[0119] The above neural network analysis module (200) can process acceleration and angular velocity data collected through the data collection module and produce walking data and motion data based thereon.

[0120] FIG. 3 is a configuration diagram showing the neural network analysis module of FIG. 1 in detail, wherein the neural network analysis module (200) may include a reliability correction unit (210) and a position correction unit (220). The reliability correction unit (210) can improve the consistency and accuracy of data by performing preprocessing, data interpolation, axis correction, outlier removal, feature scaling, and normalization of data collected from the inertial measurement device. Additionally, the position correction unit (220) can analyze changes in reliability and validity according to the attachment location of the inertial measurement device, perform repeatability and concurrent validity evaluation with existing standard data to select a location with high reliability and validity, and guide the user to the selected attachment location through the feedback provision module (400).

[0121] The present invention can achieve a level of performance similar to that of an expensive motion capture system through a reliability correction unit (210) and a position correction unit (220) while using a low-cost inertial measurement unit (IMU). The reliability correction unit maximizes the data accuracy of the sensor through data preprocessing and correction, and can secure consistent data through processes such as outlier removal, interpolation, axis correction, and feature scaling. Through such correction, data noise and measurement errors that may occur in low-cost sensors can be minimized, and more precise gait analysis can be achieved.

[0122] Furthermore, the position correction unit can enhance the reliability and validity of the data by optimizing the sensor attachment location. Since the quality of collected data may vary depending on the sensor attachment location, the position correction unit can analyze repeatability and concurrent validity to guide the user to the optimal sensor attachment location in real time. This enables the acquisition of motion data at a level identical or similar to that of high-end equipment, even with low-cost sensors.

[0123] The reliability correction unit (210) can refine acceleration and angular velocity data transmitted from the data collection module and improve the consistency and accuracy of the analysis. The module can detect and remove outliers that may occur in the collected data and maintain data continuity by interpolating missing data. In addition, it can perform axis calibration to ensure that the axis direction of each sensor is correctly aligned and apply feature scaling to ensure the range of the data is consistent. Through this preprocessing process, the reliability and validity of the data can be increased, and the analysis module can be supported in learning consistent patterns. In particular, the quality of the data is continuously monitored based on the evaluation of repeatability and concurrent validity against the existing standard (gold standard), and a correction procedure can be automatically performed if the reliability indicator falls below a threshold. Through this, the system can provide gait data and degenerative brain disease prediction results based on high accuracy and reliability.

[0124] In addition, the reliability correction unit may be characterized by using a CNN model with a data window size of 2 to 1000. Preferably, a window size of 40 to 60 may be used. More preferably, optimal data consistency and analysis performance can be optimized when a CNN model with a data window size of 50 is used, but it is not limited thereto. The type of analysis algorithm and the size of the data window may be changed through future advancements of the reliability correction unit and are not limited.

[0125] If the window size is set to less than 40, the data becomes excessively granular, potentially separating the patterns to be analyzed into fragmentary and discontinuous segments. Consequently, the neural network model may fail to properly learn the continuity and temporal dependencies of movement. Furthermore, due to the small window size, the model becomes overly sensitive to minute fluctuations in movement, increasing the risk of overfitting. This not only degrades the consistency of analysis results but also increases the likelihood of misinterpreting noise unrelated to actual walking patterns as important signals. As a result, data reliability and prediction accuracy may decline. If the window size exceeds 60, the amount of data included within the window becomes excessive, which may increase the processing burden on the model. This leads to longer computational costs and training times, and can make real-time analysis difficult. Additionally, as the window size increases, it may become difficult to detect detailed changes in movement or significant anomaly patterns. This is because the data is averaged over too wide a time range, increasing the risk of missing small changes or initial signals. As a result, early prediction of degenerative brain diseases may become inaccurate, and the sensitivity of the analysis may decrease, leading to a reduction in the overall performance of the system.

[0126] The reliability correction unit of the present invention is to be described in more detail. This is intended to clarify the technical features and feasibility of the present invention, and the scope of the present invention is not limited to these descriptions.

[0127] FIG. 3 of the present invention is a configuration diagram illustrating the neural network analysis module of FIG. 1 in detail. The reliability correction unit (210) can preprocess acceleration and angular velocity data collected from an inertial measurement unit (IMU) and perform various correction operations to increase the accuracy and reliability of the neural network analysis model. This module can play a role in increasing the consistency and accuracy of data by combining IMU data and motion capture system (VICON) data. Through the reliability correction unit, data reliability correction of the low-cost IMU sensor can be performed.

[0128] The dataset utilized by the above reliability correction unit may consist of inertial measurement unit data and motion capture system data.

[0129] The above inertial measurement device may consist of an acceleration sensor and an angular velocity sensor. The acceleration sensor can measure acceleration in each axis (x, y, z) direction, and the angular velocity sensor can measure and output angular velocity in each axis (x, y, z) direction. The data file of the above inertial measurement device is provided in CSV format and may store tilt information for each axis along with timestamps.

[0130] The above motion capture system (VICON) data is data collected using the motion capture system, which can precisely track the position of a subject in three-dimensional space using multiple high-speed cameras. The motion capture system measures the angle of rotation of the subject relative to a fixed reference coordinate system and outputs data along the (x, y, z) axes, which can be provided as an Excel file.

[0131] The data collected from the inertial measurement device and the motion capture system can be combined and organized into a structured form. The timestamp start times between the inertial measurement device and motion capture system data can be mapped and provided as an Excel file in a synchronized form. The Excel file can be used to train a deep learning model that predicts absolute angles based on the inertial measurement device data. The inertial measurement device data can be utilized as input data by using the absolute angles of the motion capture system data as labels for absolute angle prediction.

[0132] To produce the above labels, specific columns can be separated from the motion capture system file. Motion capture system files for any one or more body parts selected from the group consisting of the spine, superior angle of the scapula, rib cage, shoulder, back, knee or / and foot can be separated according to the attachment location of the inertial measurement unit. The separated labels can be combined with inertial measurement unit data and utilized as key data for training a deep learning model.

[0133] In order to accurately analyze the above dataset and train a neural network model, a data preprocessing process may be performed. The data preprocessing process may include metadata removal and axis correction, but is not limited thereto.

[0134] First, the aforementioned inertial measurement unit data may contain metadata that is not necessary for analysis. To remove this unnecessary information, the Pandas library can be used to delete the metadata and generate a refined dataset containing only the necessary information. The data generated during this process can be saved and used for future analysis.

[0135] In addition, axis correction may be performed to resolve the directional discrepancy between the inertial measurement unit data and the motion capture system data. This may occur when comparing the absolute angle obtained by integrating the inertial measurement unit data with the absolute angle of the motion capture system data, where the trends of the two data appear opposite. This may be attributed to the fact that the Z-axis of the inertial measurement unit data was recorded in a state rotated 180°. To correct this, the x and y axis data excluding the Z-axis can be multiplied by -1 to align the direction of the inertial measurement unit data with the motion capture system data.

[0136] This task can be automated by generating a Python file. The script can correct for a 180° rotation by multiplying the values ​​of all axes except the Z-axis in the inertial measurement unit data by -1. Through this correction, the orientation of the inertial measurement unit data aligns with the orientation of the motion capture system data, thereby increasing the accuracy of data analysis and neural network model training.

[0137] As described above, the preprocessed dataset generated through the metadata removal and axis correction processes is stored and can subsequently be utilized as input data for the reliability correction unit after undergoing additional preprocessing processes such as interpolation and time synchronization.

[0138] The above reliability correction unit may perform preprocessing including data interpolation, axis correction, outlier removal, feature scaling, and normalization processes, but is not limited thereto.

[0139] The reliability correction unit of the present invention may perform data interpolation and time synchronization processes to effectively combine inertial measurement device data and motion capture system data. Since inertial measurement device data and motion capture system data have different sampling levels and start times, interpolation may be essential to match them.

[0140] The above inertial measurement device data may have a sampling interval set to collect data at intervals of 0.001 seconds to 1 hour.

[0141] If the sampling interval is set to less than 0.001 seconds, an excessive amount of data is generated, which can place a significant burden on the system. As the number of data points collected per second increases rapidly, data processing and storage devices become overloaded, potentially degrading system efficiency. Furthermore, such high-speed sampling provides unnecessary precision for analyzing human movement characteristics, failing to bring substantial benefits to system performance improvement. In addition, as the amount of data increases, real-time processing in neural network analysis modules or gait analysis modules may be delayed, and noise increases due to minute sensor vibrations or electrical noise, leading to a decrease in the signal-to-noise ratio (SNR). This noise can degrade the accuracy of analysis results. Conversely, if the sampling interval exceeds one hour, subtle changes in movement may fail to be captured. Rapid movements or detailed joint motions occurring during walking are not measured, resulting in reduced data continuity and potentially requiring interpolation or correction algorithms. Such discontinuity reduces the accuracy of gait analysis or abnormal pattern detection, posing a risk of missing critical signals, particularly in cases requiring the detection of minute abnormalities such as degenerative brain diseases. In addition, if the sampling interval is prolonged, it becomes difficult to provide real-time feedback, which may result in delays in delivering immediate feedback on gait correction or balance maintenance to the user.

[0142] The motion capture system data may have a relatively low or irregular sampling rate, and linear interpolation can be performed to match the sampling rate of the inertial measurement unit data.

[0143] The resampling operation of the present invention is performed through a script written in a Python file, and the timestamps of motion capture system data can be aligned with the sampling interval of an inertial measurement unit through linear interpolation. If there are duplicate timestamps during the interpolation process, consistent data can be generated by using the average value of the group.

[0144] Since the inertial measurement unit data and the motion capture system data have different start times, time synchronization can be performed to synchronize the two datasets. To do this, the maximum value (MAX) of the first amplitude interval in the two datasets can be found to align the indices. The inertial measurement unit data can be shifted to match the start time with the motion capture system data.

[0145] Missing values ​​may occur at the end of the data after time synchronization. This is because the end times of data collection differ, and data consistency can be maintained by removing these missing values. Additionally, if outliers are present at the beginning of the data, they can be removed, and the data can be utilized starting from the flat section.

[0146] The aforementioned preprocessed data can be saved as an Excel file to verify the resampling results, but it is not limited to this. Data containing outliers can be separated to reconstruct the training dataset. This allows for the use of refined data compared to the existing data for training neural network models.

[0147] The reliability correction unit described above can perform feature scaling to increase the consistency and accuracy of data collected from the inertial measurement device. Feature scaling can improve the learning speed and enhance convergence stability by normalizing and standardizing the values ​​of the data used for model training.

[0148] The present invention can standardize each axis (x, y, z) of inertial measurement device data using a feature scale technique. Feature value standardization is performed such that each data point has a mean of 0 and a standard deviation of 1, and can be calculated using the following mathematical formula 1.

[0149] [Mathematical Formula 1]

[0150] Xstandardized=(X-μ) / σ

[0151] X: Original data value, μ: Data mean, σ: Data standard deviation

[0152] Through the above process, data with different ranges are aligned according to the same criteria, which balances the input values ​​during model training and prevents problems such as vanishing gradients or exploding gradients.

[0153] The following is a creation process to automate the above process. This process can generate a Python file to implement a function that performs scaling based on the mean and standard deviation of the data, but it is not limited to this specific program.

[0154] First, the mean and standard deviation for each axis (x, y, z) of the inertial measurement unit data can be calculated using the fit_standard_scaler function, and scaling can be performed based on this. The function performs feature standardization based on the input data and can save the resulting scaler object to a file. The saved scaler object can be reused to apply the same scaling to new data in the future.

[0155] Next, the saved scaler object can be loaded using the apply_standard_scaler function, and scaling can be applied to the new inertial measurement device data. The scaled data is stored in the same format as each column of the original data, thereby maintaining data consistency.

[0156] Performing the aforementioned scaling can improve model training speed and enhance convergence stability. Since data is standardized within a consistent range, training proceeds faster, and issues such as vanishing gradients or gradient explosions can be reduced. Furthermore, because the data range is maintained consistently, the module can learn the relationships between data points in a consistent manner, thereby improving prediction accuracy and reliability.

[0157] The above reliability correction unit can perform normalization on the ground truth data (label) to enhance the learning stability of the neural network model. Since learning may become unstable when the prediction range in a regression model is large, the model's learning process can be stabilized by standardizing the ground truth data (label).

[0158] The correct answer data (Label) used in the present invention represents an absolute angle, and its value is distributed between -180° and 180°. To normalize this to a range between -1 and 1, the following mathematical formula 2 can be used.

[0159] [Mathematical Formula 2]

[0160] Normalized Label=Label / 180 Label : Absolute angle of the correct data

[0161] Through mathematical formula 2, all absolute angles are adjusted to between -1 and 1, allowing the learning model to learn a diverse range of data uniformly. The normalized labels can be converted back to their original values ​​by multiplying by 180 in the post-processing stage to restore the values ​​predicted by the model back to their original angles.

[0162] In the reliability correction unit of the present invention, a process of applying a deep learning model to predict absolute angles by combining inertial measurement unit (IMU) data and motion capture system (VICON) data can be performed. The above process may include data preprocessing, configuration of a learning pipeline, design of a deep learning model, model training, and evaluation.

[0163] FIG. 5 is a diagram illustrating the deep learning application process according to an embodiment of the present invention.

[0164] The learning pipeline of Fig. 5 may consist of Inertial Measurement Unit (IMU) data and Motion Capture System (VICON) data. The IMU data can be used as the model's input data (X_Train), and the Motion Capture System data can be used as Y_Train, which serves as the model's correct answers (Labels). The pipeline may apply a sliding window algorithm to learn temporal patterns, which enables the learning of temporal dependencies by mapping the previous n IMU data points to a single Motion Capture System data point. The sliding window algorithm can extract a continuous time interval of the IMU data and input it into the model along with the Motion Capture System data corresponding to that interval. A Python file can be written to implement this. In the Python file, the sliding window algorithm can be defined, and the dataset can be converted into a form suitable for training.

[0165] In the present invention, two models, a CNN model and an LSTM model, can be designed to predict absolute angles. First, the CNN (Convolutional Neural Network) model has a structure advantageous for analyzing spatial features and can automatically extract patterns from input data. Since the CNN can learn consistent features without being dependent on specific locations, it can capture spatial patterns in the data. The CNN model includes a layer that converts IMU data into a 3D input format, and through this layer, the data can be transformed to fit the CNN input format. Subsequently, key patterns are extracted through two convolutional layers, and final predictions are performed through four fully connected layers. The design of the CNN model can be implemented in a Python file to extract patterns.

[0166] Furthermore, the aforementioned LSTM (Long Short-Term Memory) model is suitable for learning time-series data and sequential patterns, and can learn long-term dependencies of historical data. The LSTM continuously reflects historical data through cell states and can maintain long-term information, enabling it to learn patterns over time. The LSTM model consists of two LSTM layers; the first LSTM layer is set to return_sequences=True to return the output for all time steps, while the second LSTM layer can return the output only for the last time step. The above LSTM model learns the temporal flow of inertial measurement unit data and, based on this, can predict accurate absolute angles. The design of the LSTM model is implemented in a Python file to extract patterns.

[0167] To train the above deep learning model, you can create a train.py file, load the necessary modules, and write hyperparameters and model selection.

[0168] In addition, the Adam optimizer can be used for the optimization of the deep learning model mentioned above, but is not limited to it. The Adam optimizer requires an appropriate learning rate setting. If the learning rate is too small, the range of gradient updates becomes small, requiring a lot of time for training; conversely, if the learning rate is too large, the range of gradients becomes large, which may diverge or oscillate near the optimal point.

[0169] The performance evaluation of the above deep learning model may use Root Mean Square Error (RMSE) for numerical verification, but is not limited to this.

[0170] To run a deep learning inference model on the edge device mentioned above, model lightweighting and optimization can be performed. For example, using TensorFlow’s TFLiteConverter can convert a trained model into the TensorFlow Lite format, which can reduce memory usage and file size and increase processing speed, but is not limited to this.

[0171] The data window size of the reliability correction unit above may be 2 to 1000. More preferably, it may be 40 to 60. Even more preferably, it may be 50, but is not limited thereto.

[0172] If the window size is set to less than 40, the data becomes excessively granular, potentially separating the patterns to be analyzed into fragmentary and discontinuous segments. Consequently, the neural network model may fail to properly learn the continuity and temporal dependencies of movement. Furthermore, due to the small window size, the model becomes overly sensitive to minute fluctuations in movement, increasing the risk of overfitting. This not only degrades the consistency of analysis results but also increases the likelihood of misinterpreting noise unrelated to actual walking patterns as important signals. As a result, data reliability and prediction accuracy may decline. If the window size exceeds 60, the amount of data included within the window becomes excessive, which may increase the processing burden on the model. This leads to longer computational costs and training times, and can make real-time analysis difficult. Additionally, as the window size increases, it may become difficult to detect detailed changes in movement or significant anomaly patterns. This is because the data is averaged over too wide a time range, increasing the risk of missing small changes or initial signals. As a result, early prediction of degenerative brain diseases may become inaccurate, and the sensitivity of the analysis may decrease, leading to a reduction in the overall performance of the system.

[0173] For the learning model of the above-mentioned reliability correction unit, a CNN or LSTM model is preferred. More preferably, a CNN model may be used, but is not limited thereto.

[0174] Unlike deep learning-based CNNs or LSTMs, conventional reliability correction modules struggle to simultaneously learn complex temporal patterns and spatial characteristics. In particular, they fail to account for the continuous nature of Inertial Measurement Unit (IMU) data, which can degrade correction accuracy. Furthermore, existing algorithm-based correction modules tend to be optimized for training data, which may result in low generalization performance for new data. In contrast, CNNs and LSTMs train on large amounts of data, enabling them to enhance correction capabilities for various situations.

[0175] In addition, conventional integral-based reliability correction modules may not be able to effectively correct directional errors such as Z-axis rotation. This can lead to significant discrepancies in angle prediction and a substantial increase in RMSE values. Since CNNs and LSTMs excel at learning and correcting these directional errors, their performance can be significantly improved compared to existing modules.

[0176] The position correction unit (220) can perform the function of evaluating the reliability and validity of data based on the attachment position of the inertial measurement unit (IMU) and adjusting it to maintain the optimal attachment position. Since the attachment position of the inertial measurement unit can have a significant impact on the accuracy and consistency of the data, it may be important to monitor and correct it in real time. The position correction unit is part of a neural network analysis module and can evaluate whether the attachment status of the device is appropriate by analyzing the collected acceleration and angular velocity data.

[0177] The above position correction unit can calculate reliability and validity for data of one or more body parts selected from the group consisting of the spine (C7–T1), superior angle of the scapula, rib cage, shoulder, back, knee or / and foot, but is not limited thereto.

[0178] The above-mentioned position correction unit can first calculate the reliability and validity of the attachment position through repeatability and concurrent validity evaluations. Repeatability evaluation verifies the consistency of data repeatedly measured under the same conditions, while concurrent validity evaluation can be performed by analyzing the correlation between the data collected by the inertial measurement device and the data from the above-mentioned reliability correction unit by setting them as existing standards.

[0179] The aforementioned reliability refers to the degree of consistency observed when data is collected repeatedly under identical conditions, and repeatability and internal consistency can be analyzed to evaluate this. Repeatability can be assessed by calculating the standard deviation of data collected multiple times under the same conditions or by utilizing the coefficient of variation to verify consistency. A smaller standard deviation generally indicates higher reliability, while internal consistency is assessed by calculating the correlation coefficient between data to identify consistent patterns over time. Reliability is considered higher the closer the correlation coefficient is to 1.

[0180] The aforementioned validity indicates how accurately the measured data reflects the actual target intended for measurement. To evaluate this, concurrent validity and error assessment can be utilized. Concurrent validity assessment analyzes the correlation between inertial measurement unit (IM)-based data and existing standard data, such as motion capture systems (VICON). Additionally, the Root Mean Square Error (RMSE) can be calculated for error assessment; this RMSE is derived based on the difference between the values ​​predicted by IMPE-based data and the actual values ​​from the motion capture system. A smaller RMSE value indicates higher prediction accuracy and greater validity.

[0181] In the present invention, a correlation coefficient may be utilized to evaluate reliability and validity, and the correlation coefficient may be used as an indicator to quantify the linear relationship between two data sets, but is not limited thereto. The correlation coefficient has a value between -1 and 1, and has a value close to 1 if the two values ​​are exactly proportional, a value close to -1 if they are inversely proportional, and a value close to 0 if there is no relationship.

[0182] The above correlation coefficient can be calculated using the following method. First, the average value of the Inertial Measurement Unit (IMU) data and the Motion Capture System (VICON) data can be obtained. Then, the deviation from each data point is calculated by subtracting the average value, the product of these deviations is calculated, and all are summed. The correlation coefficient can be obtained by dividing this by the product of the sum of the squared deviations of each data point and taking the square root. The formula for the correlation coefficient is as follows: it is the value obtained by dividing the sum of the products of the deviations of each data point by the square root of the product of the sum of the squared deviations of the two data.

[0183] Based on the above evaluation results, if the reliability and validity of the attachment location fall below a threshold, the location correction module can provide a real-time guide to the user to adjust the attachment location through the feedback provision module. The threshold may be set to 0.6 to 0.8 or lower, but is not limited thereto.

[0184] If the threshold is set below 0.6, the system may allow sensor attachment locations even if their reliability and validity are low, potentially leading to a degradation in data quality. This increases the likelihood of generating incorrect gait data, which can result in inaccurate findings during gait analysis or the prediction of degenerative brain diseases. Consequently, incorrect gait patterns may be detected, unnecessary warnings may occur, and the overall reliability of the system may be compromised. Conversely, if the threshold is set above 0.8, excessive reliability and validity are required, causing the system to react overly sensitively. This necessitates frequent position adjustments even for minute changes in sensor location, potentially causing unnecessary inconvenience to the user. Furthermore, warnings may occur frequently even when there are no significant issues with the data, increasing user discomfort and potentially leading to discontinuation of system usage. Additionally, the processing load required to maintain reliability in real-time increases, leading to higher energy consumption and reduced system efficiency.

[0185] The position correction unit described above can evaluate and correct the reliability and validity of data based on the sensor attachment location in real time, thereby significantly shortening the complex skeleton modeling work previously required. Conventional technology required the generation of an accurate skeleton model and the calculation of relative positional relationships between body parts using complex algorithms for gait analysis or body movement evaluation. However, by automatically evaluating and optimizing the sensor attachment location through the position correction module, this skeleton modeling process can be simplified. Furthermore, since the position correction module supports the collection and analysis of data with the sensor attachment location optimized, highly reliable gait data can be secured without the need for separate correction work or the generation of a sophisticated skeleton model.

[0186] The above gait analysis module (300) can perform the function of analyzing the user's gait characteristics based on gait data calculated by the neural network analysis module, evaluating the degree of progression of degenerative brain disease, and detecting abnormal gait patterns. This module can perform a comprehensive gait evaluation by analyzing one or more gait indicators selected from a group consisting of joint range of motion, knee flexion angle, foot dragging, walking speed, and gait rhythm.

[0187] The gait analysis module described above can evaluate the progression of degenerative brain diseases and establish customized response and management plans based on the results. If early signs of the disease are detected through gait analysis, preventive treatment or rehabilitation programs can be initiated in collaboration with medical professionals before the disease progresses severely. This effectively slows the progression of the disease, allowing patients who are expected to develop the disease in the future or have already developed it, as well as their caregivers, to maintain their quality of life.

[0188] In addition, a personalized rehabilitation and treatment plan can be designed based on the aforementioned gait analysis data. By proposing rehabilitation exercises, physical therapy, and / or medication tailored to the abnormal gait pattern and degree of progression, treatment optimized for the patient's condition can be provided. If the joint range of motion is reduced, exercises to increase mobility may be performed, and if walking speed is reduced, a program to improve walking speed and rhythm may be applied, but is not limited thereto.

[0189] The present invention enables additional continuous management in daily life through real-time monitoring and feedback functions. Specifically, when the gait analysis module of the present invention detects an abnormal gait pattern, it provides an immediate notification, thereby assisting patients at risk of developing or who have already developed a degenerative brain disease, as well as their guardians, in preventing the worsening of the condition. Furthermore, if necessary, it recommends regular consultations with medical professionals to ensure a prompt response.

[0190] The gait analysis module described above can evaluate the effectiveness of a treatment or rehabilitation program and track the patient's progress. By periodically analyzing the gait data, it can verify whether one or more indicators selected from a group consisting of walking speed, range of motion, step count, gait cycle, presence of foot dragging, essential tremor, stride length, cadence, bilateral lower limb contact, range of motion, neck range of motion, back range of motion, and knee range of motion are improving, and continuously adjust the treatment or rehabilitation plan. This enables the provision of continuous treatment and management tailored to the patient's condition.

[0191] Sharing gait analysis data with medical professionals can enable more sophisticated and collaborative care. Along with physicians such as neurologists and rehabilitation specialists, one or more experts selected from a group consisting of physical therapists, occupational therapists, social workers, and nurses can work together to assess the patient's condition and establish a comprehensive treatment plan. This data-driven approach can enhance the efficiency of managing degenerative brain diseases and improve patients' long-term health and quality of life.

[0192] The above gait analysis module may be characterized by continuously collecting and learning real-environment data collected from actual patients with degenerative brain diseases and updating abnormal gait characteristics through a neural network model.

[0193] As an example of the above gait analysis module, the module can detect abnormal gait characteristics associated with Parkinson's disease. Parkinson's gait is generally characterized by a limited range of motion (ROM) during walking and a significant decrease in walking speed.

[0194] Accordingly, the present invention can determine that it is Parkinson's gait when the range of motion of the knee joint is limited to 10 degrees or less or the walking speed is 0.6 m / s or less when a user walks.

[0195] The above gait analysis module can provide immediate feedback if it continuously detects patterns similar to Parkinson's gait in the user's walking.

[0196] The feedback providing module (400) can perform the function of providing immediate and practical corrective feedback to the user based on abnormal gait characteristics detected by the gait analysis module or the degree of progression of degenerative brain disease. The feedback providing module can deliver various forms of feedback for gait correction and posture improvement to the user in real time, thereby helping to correct gait patterns and maintain body balance.

[0197] In addition, the feedback providing module can deliver feedback to the user in one or more ways selected from the group consisting of vibration, visual notifications, and auditory notifications. For example, if a walking pattern is detected where the user lacks knee flexion or drags their feet while walking, an immediate gait correction signal can be sent via vibration. Additionally, visual notifications can be displayed on the screens of smartphone applications, administrator platforms, and wearable devices to clearly convey the user's gait problems, and correction methods or precautions can be explained through voice guidance.

[0198] The aforementioned feedback provision module provides customized feedback based on the user's condition and gait data. It delivers various improvement guidelines in real time, such as correcting walking posture, maintaining balance, and adjusting walking speed, thereby helping the user's walking pattern to improve gradually. For example, if Parkinson's gait is consistently detected, it may provide feedback such as "Lift your knees higher" or "Be careful not to drag your feet," but is not limited to this.

[0199] Furthermore, the aforementioned feedback provision module can continuously monitor and analyze user data to track changes in gait patterns over time. This allows for the progressive optimization of feedback tailored to the user's condition, thereby providing more precise and personalized correction solutions. This functionality can continuously support the necessary gait correction for the user and help delay or prevent the progression of degenerative brain diseases.

[0200] Furthermore, if the aforementioned severe abnormal gait patterns are repeatedly detected, the feedback module can send warning notifications to administrators or medical professionals. This allows medical staff to remotely monitor the patient's condition and take prompt action if necessary.

[0201] In addition, the feedback provision module may include a function to provide real-time feedback during user-customized exercise performance. Based on the recommended BPM calculated by the neural network analysis module, it may provide one or more types of feedback selected from a group consisting of vibration, visual notifications, and auditory notifications during exercise performance.

[0202] The recommended BPM calculated above may preferably be generated based on the cadence of walking data or exercise data, but is not limited thereto. For example, the recommended BPM may be calculated by analyzing the cadence measured from the user's normal walking data, or the user may set a target BPM based on the cadence recorded during exercise.

[0203] Additionally, the feedback providing module may include a guide function that can adjust the user's sensor attachment position based on data calculated by the neural network analysis module. Based on the information provided by the position correction unit, if the sensor position is below a threshold, it may provide one or more types of feedback selected from a group consisting of vibration, visual notification, and auditory notification.

[0204] FIG. 2 is a diagram illustrating a degenerative brain disease prevention system based on an inertial measurement device with an additional content provision module and an exercise performance data collection module according to an embodiment of the present invention, wherein the system may additionally include a content provision module (500).

[0205] The above content provision module (500) can perform the function of providing customized exercise content and health management solutions based on the user's walking data and exercise performance data. This module systematically classifies and manages various content tailored to the user's condition, and can prevent the progression of degenerative brain diseases and induce improvement through continuous monitoring and analysis. FIG. 4 is a diagram illustrating the content provision module and the exercise performance data collection module of FIG. 2. The content provision module may be composed of a content management unit (510), a monitoring unit (520), a customized exercise generation unit (530), a health program unit (540), and an examination content provision unit (550), but is not limited thereto.

[0206] The above content management unit (510) can systematically classify pre-stored exercise content according to one or more criteria selected from a group consisting of body part, type, intensity, duration, and purpose, and store it in a database. This database includes exercise content suitable for various user needs and can provide one or more exercise content selected from a group consisting of gait training, balance exercises, strength training, aerobic exercises, corrective exercises, and rehabilitation exercises. The content management unit regularly updates the stored exercise content to reflect the latest exercise guidelines and training methods, and can provide exercises optimized to the user's condition changes.

[0207] The monitoring unit (520) can continuously analyze walking data and exercise performance data collected from the user and evaluate the progression and improvement status of the degenerative brain disease. Specifically, it can regularly evaluate indicators such as walking speed, joint angles, and whether there is foot dragging to provide feedback to the user regarding the improvement status. In addition, the analyzed data is summarized into daily, weekly, and monthly reports and delivered to the user and manager, enabling tracking of the disease progression status and early intervention if necessary.

[0208] The customized exercise generation unit (530) generates customized exercise content suitable for the user's condition based on the results analyzed by the monitoring unit. The customized exercise generation unit can recommend an exercise program optimized for the user's physical ability and health condition by comprehensively considering walking data and exercise performance data. For example, in the case of a patient with Parkinson's disease, walking training to improve balance and walking speed can be suggested, and in the case of a patient with early-stage Alzheimer's disease, complex exercises that stimulate cognitive function can be provided, but are not limited thereto. The customized exercise content is designed to match the individual user's goals and characteristics, and the difficulty level can be adjusted step by step to maximize continuous participation and effectiveness.

[0209] The customized exercise generation unit analyzes the cadence calculated by the neural network analysis module to calculate a BPM recommended to the current user, and can provide the calculated BPM as exercise content combined with one or more feedbacks selected from a group consisting of vibration, visual notifications, and auditory notifications. For example, if the calculated cadence is 50, it can provide exercise content that provides one or more feedbacks selected from a group consisting of vibration, auditory notifications, and visual notifications of 60 BPM, but is not limited thereto.

[0210] The above health program unit (540) can provide a customized lifestyle improvement protocol by analyzing one or more values ​​selected from a group consisting of the user's walking data, exercise performance data, and the degree of progression of degenerative brain disease. The lifestyle improvement protocol may consist of one or more elements selected from a group consisting of diet, sleep management, meditation, psychological counseling, customized nutritional supplements, walking, and running. For example, a user with reduced balance may be recommended vitamin D supplements and balance maintenance exercises, and a user with poor sleep quality may receive a meditation program for sleep improvement and advice on sleep habits, but is not limited to these. Such a customized protocol can help improve the user's overall health and lifestyle habits and reduce the risk of degenerative brain disease.

[0211] The above-mentioned test content providing unit (550) can provide various test contents for evaluating a user's degenerative brain disease. The above-mentioned test content providing unit may consist of simple tests and precise tests, but is not limited thereto.

[0212] The above-mentioned simplified test may be provided in a digitized form of the Timed Up and Go test (TUG), which was developed to assess the risk of Parkinson's disease, dementia, and falls. Upon receiving a test start signal, the user may perform the procedure of standing up from a chair, moving 3 meters, returning to the original position, and sitting back down in the chair. If the test time is 30 seconds or longer, the risk of Parkinson's disease, dementia, and falls may be assessed as high.

[0213] The aforementioned precise examination can be provided based on real-world data collected from actual patients with degenerative brain diseases. The examination content provider analyzes the user's gait data through a neural network analysis model and can detect abnormal gait patterns. The data collected during this process is continuously updated, and the system reflects changes in the data to enable more accurate diagnosis and prediction.

[0214] The aforementioned collected exercise data can be systematically collected, analyzed, classified, and stored by a data collection module. The data collection module measures acceleration and angular velocity data in real time, and based on this, the neural network analysis module analyzes the patterns and characteristics of the exercise performed by the user to convert and store them as quantified exercise data.

[0215] The above data collection module can continuously record data of customized exercises performed by the user through the content provision module. When performing specific gait training or balance maintenance exercises, the data collection module can measure changes in acceleration and angular velocity occurring during each exercise movement. The measured data can be processed by the neural network analysis module and converted into one or more detailed exercise data selected from a group consisting of walking speed, joint amplitude range, step count, gait cycle, presence of foot dragging, essential tremor, stride length, cadence, bilateral lower limb contact, joint range of motion, neck joint range of motion, back joint range of motion, and knee joint range of motion.

[0216] The exercise data generated above is classified and stored as exercise data in the data collection module. The data can be classified according to one or more criteria selected from a group consisting of exercise body parts, purpose, type, duration, intensity, and repetition count, and the trends and changes in the user's exercise performance can be analyzed through the data accumulated over the long term. Furthermore, the data collection module can support the continuous optimization of customized exercise content by evaluating the degree of improvement based on the user's exercise performance records and providing feedback to the content provision module.

[0217] The aforementioned exercise data can be linked with a monitoring unit and periodically reported to users and administrators. This allows users to intuitively check their exercise performance and improvement status, while enabling medical professionals or administrators to assess the progression of degenerative brain diseases and take appropriate measures. If a pattern of continuous improvement in walking speed or an increase in balance maintenance time appears, the exercise program can be evaluated as effective; conversely, if signs of degeneration persist, notifications indicating the need for gait correction or additional training may be provided, but are not limited to these cases.

[0218]

[0219] In addition, the present invention provides a method for providing information to optimize the system. Each step can be organically linked to maximize the efficiency and accuracy of the system.

[0220]

[0221] The present inertial measurement device-based method for providing information on the prevention of degenerative brain diseases may provide an inertial measurement device-based degenerative brain disease prevention system characterized by comprising: a data collection step of measuring acceleration and angular velocity data of an inertial measurement device attached to a user's body part in real time and collecting said measured data; a neural network analysis step of analyzing the data collected through said data collection module using a neural network analysis model to produce gait data; a gait analysis step of analyzing said gait data to evaluate the degree of progression of degenerative brain disease and detect abnormal gait characteristics; and a feedback provision step of providing real-time feedback when said abnormal gait characteristics are detected.

[0222] The above system may additionally include a content provision step that provides customized exercise content tailored to the user's condition prior to the data collection step.

[0223] First, the data collection step described above can collect user movement data in real time using an Inertial Measurement Unit (IMU) sensor. The IMU sensor is attached to a suitable number of body parts selected from the group consisting of specific positions of the user's spine (C7~T1), the superior angle of the scapula, the rib cage, the shoulder, the back, the knee or / and the foot, and can accurately measure 3-axis acceleration and angular velocity data of the IMU. The sensor used in the present invention is not limited to an IMU but can be implemented as a sensor of various forms and performance.

[0224] The collected data is grouped into data collection time intervals of 0.001 seconds to 1 hour, processed in the form of a data window, and can be transmitted to a data processing device. This data provides biomechanical information of the user, such as joint angles and movement speeds, and can be used as basic data for analyzing movement patterns.

[0225] In addition, to maintain data continuity and consistency, the data collection stage can organize the measured data into a consistent format through time synchronization and transmit it to the neural network analysis module.

[0226] The neural network analysis module of the present invention can produce walking data by analyzing the data collected through the data collection step using a neural network analysis model.

[0227] Specifically, the neural network analysis step processes data received from the data collection module using deep learning-based analysis technology, and the analysis step may include a reliability correction step that corrects the reliability of gait data through a neural network model trained based on existing standard data measured from a plurality of conventional inertial measurement devices and a motion capture system. The reliability correction step can improve the consistency and accuracy of the data by preprocessing the data collected from the inertial measurement devices and performing data interpolation, axis correction, outlier removal, feature scaling, and normalization.

[0228] In addition, the neural network analysis step may analyze data by adopting a CNN model with a data window size of 2 to 1000, and more preferably, may analyze with a CNN model with a data window size of 40 to 60. The most preferred data window size of the present invention may be set to 50, but is not limited thereto.

[0229] Additionally, the neural network analysis step may include a position correction step that, after the reliability correction step, analyzes changes in reliability and validity according to the attachment location of the inertial measurement device, performs repeatability and concurrent validity evaluations with existing standards to select a location with high reliability and validity, and guides the selected attachment location to the user through the feedback provision step.

[0230] Specifically, after the reliability correction step, if the reliability and validity for each attachment location are evaluated and the optimal attachment location is determined, the position correction step can sequentially guide the user to the corresponding location information through the feedback provision step and readjust the sensor attachment location if necessary.

[0231] The above position correction step; can provide a guide to the user to adjust the sensor attachment position in real time through the above feedback providing step; when the reliability and validity of the inertial measurement device drop below a threshold. The threshold may be 0.6 to 0.8 or less. More preferably, it may be set to 0.7, but is not limited thereto.

[0232] The above neural network analysis step can analyze data collected from an inertial measurement device to calculate one or more gait data selected from a group including walking speed, joint amplitude range, number of steps, gait cycle, presence of foot dragging, essential tremor, stride length, cadence, contact period of both lower limbs, range of motion of joints, range of motion of neck joints, range of motion of back joints, and range of motion of knee joints, and more preferably, may include knee flexion angle as a key indicator, but is not limited thereto.

[0233] The aforementioned gait data can play a crucial role in detecting early signs of disease by precisely analyzing the user's movement patterns. In particular, the neural network analysis step performs a process of removing outliers and standardizing the data to increase data reliability, thereby enhancing the reliability and accuracy of the analysis results.

[0234] The above gait analysis step can analyze the gait data to evaluate the degree of progression of degenerative brain disease and detect abnormal gait characteristics.

[0235] In addition, the gait analysis step described above can continuously collect and learn real-environment data collected from actual patients with degenerative brain diseases and update abnormal gait characteristics through a neural network model.

[0236] The above feedback provision step may provide feedback to the user through one or more selected from a group consisting of vibration, visual notifications, and auditory notifications according to abnormal signs in the gait data calculated by the gait analysis module, but is not limited thereto.

[0237] The above feedback may be characterized by being able to improve one or more items selected from a group consisting of gait correction, posture correction, balance maintenance, and walking speed control.

[0238] The above feedback provision step can continuously monitor the user's walking data to progressively optimize personalized feedback based on walking patterns and changes analyzed over time.

[0239] The present invention may further include a content provision step that provides customized exercise content tailored to the user's condition prior to the data collection step.

[0240] The above-mentioned content provision step can perform the function of providing customized exercise content and health management solutions based on the user's gait data and exercise performance data. The above-mentioned content provision step systematically classifies and manages various content tailored to the user's condition, and can prevent the progression of degenerative brain diseases and induce improvement through continuous monitoring and analysis. The above-mentioned content provision step may consist of a content management step, a monitoring step, a customized exercise generation step, a health program step, and a test content provision step.

[0241] The above content provision step may include a content management step of systematically classifying pre-stored exercise content, storing it in a database according to one or more criteria selected from a group consisting of the type, intensity, duration, and purpose of each content, and managing and updating it.

[0242] The above content provision step may include a monitoring step of continuously collecting and analyzing the gait data and exercise data, evaluating the progression status and improvement of the degenerative brain disease, and regularly reporting to the user.

[0243] In addition, the content provision step may include a customized exercise generation step of analyzing the user's walking data and exercise data and providing customized exercise content suitable for the user from among the exercise content stored in the content management unit according to the analyzed result.

[0244] The above customized exercise generation step analyzes the cadence calculated by the neural network analysis module to calculate a BPM recommended to the current user, and can provide the calculated BPM as exercise content combined with one or more feedbacks selected from a group consisting of vibration, visual notifications, and auditory notifications.

[0245] The above content provision step may include a health program step of providing one or more customized protocols selected from a group consisting of a diet for lifestyle improvement, sleep management, meditation, psychological counseling, customized nutritional supplements, walking, and running, based on one or more pieces of information selected from a group consisting of the degree of progression of the user's degenerative brain disease, the walking data, and the exercise data.

[0246] The above-mentioned content provision step may include a step of providing various test content for the evaluation of a user's degenerative brain disease. The above-mentioned test content provision step may consist of simple tests and precise tests, but is not limited thereto.

[0247] The customized exercise data provided through the above-mentioned content provision step can be systematically collected, analyzed, classified, and stored in the data collection step. The data collection step can measure acceleration and angular velocity data in real time, analyze the data in the neural network analysis step to convert it into quantified exercise data, and store it in the data collection module.

[0248] FIG. 7 is a schematic diagram illustrating an inertial measurement device of the present invention. In one embodiment of the present invention, the inertial measurement device may include an acceleration sensor and an angular velocity sensor, but is not limited thereto.

[0249] The above acceleration sensor can measure the user's movement occurring in the three axis directions of X, Y, and Z, and the above angular velocity sensor can precisely capture the user's joint rotation or body movement by measuring the angular velocity occurring in the three axis directions of X, Y, and Z.

[0250] In another embodiment, the inertial measuring device may further include additional sensors of a geomagnetic sensor or strain gauge as needed, but is not limited thereto.

[0251] Meanwhile, the shape of the inertial measuring device shown in FIG. 7 is merely one embodiment of the present invention, and within the technical scope of the present invention, the shape, size, and internal configuration of the device can be varied in various ways. For example, it may be manufactured in the form of a miniature wearable or implemented as a simple box-shaped housing, but is not limited thereto. Such variations should also be understood to be included within the scope of the present invention.

[0252] FIG. 8 is a diagram illustrating an exemplary cradle for stably attaching the inertial measuring device of the present invention to a part of a user's body.

[0253] In one embodiment of the present invention, the cradle may include a Velcro strap to allow easy attachment to the user's knees, shoulders, and back, but is not limited thereto. Additionally, the main body of the cradle may be formed of an elastic material or lightweight plastic to support the stable fixation of the inertial measurement device even during walking or running, but is not limited thereto.

[0254] As shown in FIG. 8, the inertial measuring device may be inserted into the cradle or secured using a coupling part, and may be designed to be easily detachable for sensor position correction or maintenance and repair work on the inertial measuring device if necessary, but is not limited thereto.

[0255] In another embodiment of the present invention, unlike the cradle (800) shown in FIG. 8, it can also be implemented in the form of a wearable band, a compression sleeve, or an insole inside a shoe.

[0256] The present invention enables early diagnosis of degenerative brain diseases using an inertial measurement device. The method includes a data collection step of measuring acceleration and angular velocity data of an inertial measurement device attached to a user's body part in real time and collecting the measured data.

[0257] Subsequently, it includes a neural network analysis step that calculates walking data by analyzing the data collected through the data collection module using a neural network analysis model.

[0258] In addition, it includes a gait analysis step that analyzes the gait data to evaluate the degree of progression of degenerative brain disease and detects abnormal gait characteristics, and additionally includes a feedback provision step that provides real-time feedback when abnormal gait characteristics are detected.

[0259] In addition, through the aforementioned gait analysis, minute indicators such as the user's walking speed, stride length, cadence, and contact points of both legs are tracked in real time and compared with normal patterns learned by a neural network model to preemptively identify early signs of Parkinsonian gait or balance impairment, and based on this, early diagnosis of degenerative brain diseases is performed.

[0260]

[0261] The present invention will be explained in more detail below through examples. These examples are intended to explain the invention more specifically, and the scope of the invention is not limited to these examples.

[0262]

[0263] Example 1. LSTM Model Training

[0264] The LSTM model was trained using a reliability correction unit. Data was collected at a sampling interval of 0.01 seconds, and interpolation and resampling were performed at 0.01-second intervals to maintain data consistency. Through this data preprocessing, training data with uniform time intervals was constructed. Subsequently, the data window sizes were set to 10 and 50 to train the LSTM models. Through this, the pattern learning effect across different time ranges was compared and performance evaluated, which is shown as a graph in Figure 6a.

[0265] As a result, according to Figure 6a, performance differences based on the data window size were confirmed during the training of the LSTM model. In all graphs, the loss showed a decreasing trend as the number of training iterations (epochs) progressed, confirming that training was proceeding smoothly. In the case of the model with a data window size of 10, the loss continuously decreased on the training data, but on the validation data, the loss showed an increasing trend after the 151st and 185th training iterations, suggesting that overfitting is occurring. On the other hand, the model with a data window size of 50 showed lower loss values ​​on the validation data and demonstrated relatively superior performance.

[0266]

[0267] Example 2. CNN Model Training

[0268] The CNN model was trained using a confidence correction unit. Data was collected at a sampling interval of 0.01 seconds, and interpolation and resampling were performed at 0.01-second intervals to maintain data consistency. Through this data preprocessing, training data with uniform time intervals was constructed. Subsequently, the data window sizes were set to 10 and 50 to train the CNN models. Through this, the pattern learning effect across different time ranges was compared, and the performance was evaluated and presented as a graph in Figure 6b.

[0269] As a result, according to Figure 6b, a trend of decreasing loss was observed in all graphs as the number of training iterations (epochs) progressed during CNN model training. This indicates that the model training proceeded smoothly. It showed lower loss compared to the LSTM model in Figure 6a, which means that the CNN model has higher reliability and validity than the LSTM model. In particular, the model with a window size of 50 showed lower loss in the validation data, which means that its performance is relatively superior to that of the model with a window size of 10. Furthermore, seeing that the training loss and validation loss continued to decrease, it was confirmed that there is potential for further model training.

[0270]

[0271] Comparative Example 1. IMU data using conventional integration

[0272] The model performance was evaluated using an algorithm that converts IMU data into absolute angles using conventional integration. The results of estimating angles by simply integrating data collected from an inertial measurement unit (IMU) were evaluated using RMSE values, and the results are shown in Table 1.

[0273] The RMSE values ​​were recorded for the data classified into K_HIP, K_KNEE, and Y_HIP1. K_HIP represents the hip joint angle and evaluates the accuracy of the hip joint angle predicted by the model. K_KNEE represents the knee joint angle and is an indicator evaluating the prediction performance regarding knee movement. Additionally, Y_HIP1 is a value related to the right hip joint angle, allowing verification of the hip joint movement prediction performance.

[0274]

[0275] Data Name K_HIPK_KNEEY_HIP1 Conventional Integral RMSE Value 85.2132854.4777448.6039

[0276] According to Table 1, the method of estimating angles by integrating conventional IMU data showed a higher RMSE value compared to Examples 3 and 4 below. For the K_HIP data, the RMSE value was recorded at 85.21328, which is approximately 19.8 times higher than the RMSE value of 4.3139 of the CNN model with Z-axis correction applied in Table 2. For the K_KNEE data, the RMSE value was 54.47774, showing an error approximately 22.6 times higher compared to 2.4071 of the CNN model with Z-axis correction applied.

[0277] In the Y_HIP1 data as well, the conventional integration method recorded an RMSE value of 48.6039, which is approximately 19.5 times higher than the CNN model's 2.4888 after Z-axis correction. These results demonstrate that the conventional integration-based IMU data processing method is vulnerable to cumulative error.

[0278]

[0279] Example 3. Training model after Z-axis rotation

[0280] CNN and LSTM models were trained using a reliability correction unit. Data was collected at a sampling interval of 0.01 seconds, and interpolation and resampling processes were performed at 0.01-second intervals to maintain consistency. Through the above data preprocessing, training data with uniform time intervals was constructed. In Example 3 above, during the data preprocessing process of the reliability correction unit, training was performed after correcting the Z-axis direction by multiplying all axes except the Z-axis by -1 and rotating them 180 degrees.

[0281] To train the CNN and LSTM models, the data window sizes were set to 10 and 50, respectively, to compare the pattern learning effects across different time ranges and evaluate performance. The performance of each model was evaluated using RMSE values, and the results are shown in Table 2.

[0282] The RMSE values ​​were recorded for the data classified into K_HIP, K_KNEE, Y_HIP1, and Y_HIP2. K_HIP represents the hip joint angle and evaluates the accuracy of the hip joint angle predicted by the model. K_KNEE represents the knee joint angle and is an indicator evaluating the prediction performance regarding knee movement. Additionally, Y_HIP1 represents the right hip joint angle and Y_HIP2 represents the left hip joint angle, allowing for verification of the prediction performance for the movement of both hip joints.

[0283] K_HIPK_KNEEY_HIP1Y_HIP2 Window 10 CNN model RMSE value 5.6916 4.8367 3.2525 2.7998 Window 50 CNN model RMSE value 4.3139 2.4071 2.4888 1.8923 Window 10 LSTM model RMSE value 7.0775 5.1625 5.9499 3.8782 Window 50 LSTM model RMSE value 5.1758 4.3143 3.1576 3.1443

[0284] According to Table 2, the CNN model with a window size of 50 recorded the lowest RMSE value overall, confirming that the CNN model demonstrated higher performance after Z-axis rotation correction was applied. For the K_HIP data, the CNN model with a window size of 50 showed an RMSE value of 4.3139, which is approximately 24.2% lower than the RMSE value of 5.6916 of the CNN model with a window size of 10. In the K_KNEE data as well, the CNN model with a window size of 50 showed an RMSE value of 2.4071, which is approximately 50.2% lower than the 4.8367 of the CNN model with a window size of 10.

[0285] On the other hand, the LSTM model showed higher error than the CNN model regardless of the window size. In the K_HIP data, the LSTM model with a window size of 50 had an RMSE value of 5.1758, which was about 19.9% ​​higher than the CNN model with a window size of 50, which had an RMSE of 4.3139. Additionally, in the Y_HIP2 data, the CNN model with a window size of 50 recorded an RMSE value of 1.8923, while the LSTM model under the same conditions recorded 3.1443, showing an error about 66.2% higher than the CNN model.

[0286] These results suggest that CNN models effectively learn spatial patterns, and their performance is superior, particularly when the orientation of the data is improved through Z-axis rotation correction. On the other hand, while LSTM models are specialized in learning temporal patterns, they showed relatively high errors even with Z-axis correction applied. Therefore, the present invention confirmed that CNN models with Z-axis correction applied provide the most reliable results in predicting hip and knee joint angles, and that accuracy can be improved by up to 66% compared to LSTM models.

[0287]

[0288] Example 4. Training model without Z-axis rotation

[0289] CNN and LSTM models were trained using a reliability correction unit. Data was collected at a sampling interval of 0.01 seconds, and interpolation and resampling processes were performed at 0.01-second intervals to maintain consistency. Through the above data preprocessing, training data with uniform time intervals was constructed. In Example 4 above, training was conducted using the original data as is, without applying Z-axis rotation correction during the data preprocessing of the reliability correction unit.

[0290] To train the CNN and LSTM models, the data window sizes were set to 10 and 50, respectively, to compare the pattern learning effects across different time ranges and evaluate performance. The performance of each model was evaluated using RMSE values, and the results are shown in Table 3.

[0291] The RMSE values ​​were recorded for the data classified into K_HIP, K_KNEE, Y_HIP1, and Y_HIP2. K_HIP represents the hip joint angle and evaluates the accuracy of the hip joint angle predicted by the model. K_KNEE represents the knee joint angle and is an indicator evaluating the prediction performance regarding knee movement. Additionally, Y_HIP1 represents the right hip joint angle and Y_HIP2 represents the left hip joint angle, allowing for verification of the prediction performance for the movement of both hip joints.

[0292] K_HIPKNEEHIP1HIP2 Window 10 CNN model RMSE value 6.23755.00034.02113.5862 Window 50 CNN model RMSE value 5.50933.36773.99833.5474 Window 10 LSTM model RMSE value 6.74566.65955.68877.4133 Window 50 LSTM model RMSE value 5.63506.66405.39925.5217

[0293] According to the results in Table 3, the CNN model with a window size of 50 showed the lowest RMSE values ​​overall, even without applying Z-axis correction. In the K_HIP dataset, the CNN model with a window size of 50 recorded an RMSE value of 5.5093, which is approximately 11.7% lower than the 6.2375 of the CNN model with a window size of 10. In the K_KNEE dataset as well, the CNN model with a window size of 50 recorded an RMSE value of 3.3677, which is approximately 32.7% lower than the 5.0003 of the model with a window size of 10. In the Y_HIP2 dataset, the CNN model with a window size of 50 recorded 3.5474, showing a slight improvement over the 3.5862 of the model with a window size of 10.

[0294] Compared to Comparative Example 1 above, the CNN model recorded a lower RMSE value than the conventional integral-based IMU data processing method, even without applying Z-axis correction. In the K_HIP data, Comparative Example 1 had an RMSE value of 85.21328, while the CNN model showed a performance decrease of approximately 93.5% to 5.5093. In the K_KNEE data, the CNN model decreased by approximately 93.8% to 3.3677 compared to the RMSE value of 54.47774 of Comparative Example 1, and in the Y_HIP1 data as well, the CNN model recorded a value of approximately 91.8% lower to 3.9983 compared to 48.6039 of Comparative Example 1.

[0295] However, compared to Example 3, the error increased when Z-axis correction was not applied. For the K_HIP data, the CNN model with a window size of 50 in Example 3, which applied Z-axis correction, had an RMSE value of 4.3139, whereas without Z-axis correction, it was 5.5093, showing an error approximately 27.7% higher. In the K_KNEE data as well, the RMSE value of the CNN model with Z-axis correction applied was 2.4071, while without correction, it was 3.3677, showing a value approximately 39.9% higher. In the Y_HIP2 data, the CNN model with Z-axis correction applied recorded an RMSE value of 1.8923, whereas without correction, it was 3.5474, showing an error approximately 87.5% higher. This demonstrates that Z-axis correction plays an important role in improving model performance by improving data orientation.

[0296] As a result of a comprehensive analysis of Examples 3, 4, and Comparative Example 1, it was clearly confirmed that the effects of whether reliability correction was applied, the type of deep learning model (CNN and LSTM), the setting of the window size, and Z-axis correction on prediction performance were clearly identified. An advanced comparative analysis of each element is as follows.

[0297] In the conventional Comparative Example 1, angles were estimated by simply integrating IMU data, and as a result, very high RMSE values ​​were observed due to accumulated error. The RMSE values ​​for K_HIP were 85.21328, K_KNEE was 54.47774, and Y_HIP1 was 48.6039. This represents an error up to approximately 22.6 times higher than that of CNN and LSTM models with confidence correction applied. Consequently, the conventional method clearly had limitations in real-time motion prediction.

[0298] On the other hand, in Examples 3 and 4, where reliability correction was applied, the error was significantly reduced. In particular, the CNN model recorded RMSE values ​​of approximately 4.3–5.5 for K_HIP, approximately 2.4–3.3 for K_KNEE, and approximately 1.9–4.0 for Y_HIP1 and Y_HIP2 through reliability correction, demonstrating much more precise prediction performance than the conventional method. This proves that the reliability correction algorithm is effective in improving the cumulative error problem and increasing data accuracy.

[0299] As a result of comparing CNN and LSTM models, the CNN model showed strengths in spatial pattern learning. In Example 3, the CNN model recorded the lowest error with RMSE values ​​of K_HIP 4.3139, K_KNEE 2.4071, Y_HIP1 2.4888, and Y_HIP2 1.8923.

[0300] On the other hand, while the LSTM model is specialized for learning temporal patterns, it had limitations in learning spatial characteristics. In Example 3, the RMSE values ​​of the LSTM model were K_HIP 5.1758, K_KNEE 4.3143, Y_HIP1 3.1576, and Y_HIP2 3.1443, showing an error that was approximately 19.9% ​​to 66.2% higher than that of the CNN model. This implies that the CNN model is more suitable when dealing with spatial data such as real-time body movements.

[0301]

[0302] As a result of comparing model performance when the window size was set to 10 and 50, both CNN and LSTM models showed improved performance at a window size of 50. For the CNN model, the K_HIP test recorded an RMSE of 4.3139 with a window size of 50, a decrease of approximately 24.2% compared to 5.6916 with a window size of 10. In K_KNEE as well, the error decreased by approximately 50.2% with a window size of 50, recording 2.4071 compared to 4.8367 with a window size of 10. This indicates that the model is efficient when the window size is 50.

[0303] Comparing Example 3, which applied Z-axis correction, with Example 4, which did not, it was confirmed that Z-axis correction has a significant impact on prediction performance. In Example 3, which applied Z-axis correction, the RMSE values ​​of the CNN model recorded the lowest figures, with K_HIP 4.3139, K_KNEE 2.4071, Y_HIP1 2.4888, and Y_HIP2 1.8923. On the other hand, in Example 4, which did not apply Z-axis correction, the RMSE values ​​were K_HIP 5.5093, K_KNEE 3.3677, Y_HIP1 3.9983, and Y_HIP2 3.5474, showing an increase in error of approximately 20% to 40%. This implies that the model can effectively learn spatial patterns by aligning the directionality of the data through Z-axis correction, and demonstrates that if correction is not performed, prediction performance deteriorates due to data inconsistency.

[0304] As a result of comprehensively analyzing Examples 3, 4, and Comparative Example 1 of the present invention, the CNN model (Example 3) with reliability correction and Z-axis correction applied showed precise and reliable performance, with an RMSE value up to 22.6 times lower than the conventional integration method (Comparative Example 1). In particular, the CNN model has strengths in learning spatial patterns and showed superior performance compared to the LSTM model in predicting joint angles. Additionally, when the window size was set to 50, more patterns could be learned, leading to an overall improvement in model performance. Z-axis correction significantly improved prediction performance by aligning the directionality of the data, and it was confirmed that the error increased by approximately 20% to 40% when this was not applied.

[0305]

[0306] The system of the present invention, based on the organic operation of the reliability correction unit and the position correction unit, can analyze data collected from an inertial measurement unit (IMU) in a highly refined and optimized state. The reliability correction unit can maximize the consistency and accuracy of sensor data through processes such as data preprocessing, outlier removal, axis correction, and feature scaling, while the position correction unit can support maintaining the optimal sensor position by evaluating and correcting the reliability and validity according to the sensor's attachment location in real time.

[0307] The above reliability correction unit and position correction unit operate organically, and at the same time, multiple inertial measurement units (IMUs) can be used to measure accurate data in real time from various body parts. More preferably, by attaching inertial measurement units to the lower and upper parts of the knee, respectively, the movement trajectory and rotation angle of the knee joint can be precisely captured, but is not limited thereto. The data collected through this can secure reliability and precision similar to existing high-cost motion capture systems.

[0308] Through data analysis, the accuracy of predicting gait patterns and joint angles can be improved compared to the conventional method of estimating angles via integration. Furthermore, by effectively resolving the issue of error accumulation in sensor data, body movements can be precisely evaluated in real time. Additionally, by integrating and analyzing data collected from multiple sensors attached to various body parts, the movement connectivity and balance status between the upper and lower body can be comprehensively assessed, enabling the provision of user-customized feedback and corrective solutions.

[0309] In addition, the neural network analysis module can further enhance analysis performance by utilizing deep learning models such as CNN and LSTM to simultaneously learn spatial and temporal patterns and aligning the directionality of the data through Z-axis rotation correction. Based on this refined data, the gait analysis module can detect early signs of degenerative brain diseases and evaluate abnormal gait characteristics in real time, and provide immediate and customized feedback to the user in conjunction with the feedback provision module.

[0310]

[0311] The detailed description for carrying out the invention disclosed above is provided to enable those skilled in the art to implement and practice the invention. Although the above description refers to preferred principal modules of the invention, those skilled in the art will understand that various modifications and changes can be made to the invention without departing from the scope of the invention. For example, those skilled in the art may utilize each configuration described in the above embodiments in a manner that combines them with one another.

[0312] Accordingly, the present invention is not intended to be limited to the embodiments shown herein, but to be given the broadest scope consistent with the principles and novel features disclosed herein.

[0313] The present invention may be embodied in other specific forms without departing from the spirit and essential features of the invention. Accordingly, the above detailed description should not be interpreted restrictively in all respects but should be considered exemplary. The scope of the invention shall be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the invention are included within the scope of the invention. The invention is not intended to be limited to the embodiments shown herein, but to be given the broadest possible scope consistent with the principles and novel features disclosed herein. Furthermore, embodiments may be constructed by combining claims that are not explicitly related in the claims, or by including them as new claims through amendments made after filing.

[0314] The present invention will be explained in more detail below with specific examples. However, the embodiments described below are provided as examples to ensure that the concept of the present invention is sufficiently conveyed to those skilled in the art.

[0315] Accordingly, the present invention is not limited to the embodiments presented below and may be embodied in other forms, and the embodiments presented below are described merely to clarify the concept of the present invention and are not limited thereto.

[0316]

[0317] definition

[0318] Expressions such as “comprising,” “comprising,” “having,” etc. as used herein should be understood as open-ended terms implying the possibility of including other embodiments in a manner similar to “comprising,” unless otherwise stated in the phrase or sentence containing such expressions.

[0319] The term "and / or" as used herein may mean any one or more of the items, any combination of the items, or all of the items in relation to the term.

[0320] As used herein, the term “output” may refer to a process of generating or deriving a result by processing or calculating input data. This includes the act of obtaining a specific result value through one or more data processing methods selected from the group consisting of data analysis, calculation, or algorithmic processing, and is not necessarily limited to a specific format or method.

[0321] As used herein, the term "generation" may refer to a process of creating new data, materials, structures, or results based on specific input values ​​or conditions. This may be accomplished in various ways, such as physical, chemical, or data processing, and is not necessarily limited to a specific methodology or procedure.

[0322] As used herein, the term "suitable" may refer to a state that conforms to a specific purpose or situation or satisfies requirements. This may be evaluated based on one or more criteria selected from the group consisting of physical characteristics, data reliability, and device performance, and is not necessarily limited to a specific format or value. In the present invention, the term "suitable" is used in the context of the attachment location of the inertial measurement device, the data processing method, or the reliability of analysis results, and may refer to satisfying the conditions necessary to ensure the efficiency and accuracy of the system. Furthermore, "suitable" is a concept that may vary depending on the environment, purpose of use, or application conditions, and may be interpreted within the scope of purpose / specific realization for solving the technical problem of the present invention, including all such variations and possibilities of application.

[0323] As used in this specification, the term “location information” refers to user location coordinate data calculated through a satellite-based positioning device (GNSS: GPS, GLONASS, Galileo, BeiDou) and may include one or more pieces of information selected from the group consisting of absolute coordinates, continuous change in position, movement speed, direction of movement, and elevation information. Location information may vary depending on signal strength and surrounding environment and is not limited to a specific measurement method or coordinate system.

[0324] As used herein, the term “Inertial Measurement Unit (IMU)” refers to a device that measures a user’s linear acceleration and rotational angular velocity, including an acceleration sensor and a gyroscope. The IMU may include sensor configurations such as 3-axis or 6-axis, mounting positions, sizes, and package forms, and is not limited to a specific structure or form.

[0325] As used in this specification, the term “drift” refers to a phenomenon in which the deviation or error of a reference value, which accumulates in acceleration and angular velocity data measured over a long period by an inertial measurement unit (IMU), gradually increases over time. This drift may be caused by sensor bias, temperature changes, mechanical micro-deformation, changes in attachment position, or internal sensor noise, and is characterized by the sensor output slowly shifting in a direction that does not match the actual movement. Therefore, to correct this drift, the present invention performs a reliability correction procedure that includes axis alignment correction, bias removal, adaptive filtering, and signal re-normalization processes.

[0326] The term “bias” as used in this specification refers to a characteristic in which an IMU sensor outputs a constant offset value in a specific axis direction even when stationary, and said bias can be a major cause of drift.

[0327] As used in this specification, the term “axis alignment correction” refers to a transformation process that corrects tilt or rotational errors of sensor axes so that the sensor measurement axes of an inertial measurement unit (IMU) coincide with the actual spatial coordinate system. Axis alignment correction may be performed using one or more methods selected from the group consisting of rotation matrices, direction cosine matrices (DCM), and quaternion transformations, thereby ensuring that acceleration and angular velocity data are aligned with the body reference axis or the ground reference axis. The present invention may include axis alignment correction to maintain the accuracy of gait analysis and GNSS-IMU fusion, and the specific method is not limited to a specific method.

[0328] As used herein, the term “acceleration data” refers to linear acceleration values ​​in the X-axis, Y-axis, and Z-axis directions measured by an inertial measurement device. It may include data signals continuously collected in real time and is not limited to a specific sampling rate or preprocessing method.

[0329] As used herein, the term “angular velocity data” refers to angular velocity information of rotational motion obtained through a gyroscope. It may include rotational values ​​of the roll, pitch, and yaw axes and is not limited to a specific unit system or filtering method.

[0330] As used herein, the term “gait data” refers to gait-related characteristic values ​​calculated through the analysis of temporal patterns of acceleration and angular velocity data. It may include various data such as walking speed, essential tremor and bilateral ground contact, gait cycle, cadence, stride length, rhythm, balance, and / or foot dragging, and is not limited to specific analysis algorithms or event detection methods.

[0331] As used herein, the term "cadence" may refer to a biomechanical parameter representing the frequency of periodic movement over a specific period of time, particularly the number of steps per minute in a user's gait cycle. This serves as key data for quantitatively evaluating a user's movement characteristics and can be utilized for gait analysis, evaluation of movement patterns, and / or early prediction of degenerative brain diseases. The "cadence" is not necessarily limited to walking and can also be applied when measuring or evaluating repetitive movement patterns of the body. Specifically, in one embodiment of the present invention, "cadence" can be used to precisely analyze a user's walking speed, rhythm, and pattern based on data collected from an Inertial Measurement Unit (IMU). It can serve as a key analysis variable in a neural network analysis module and be utilized as an important factor for detecting signals of degenerative brain diseases or identifying gait abnormalities.

[0332] As used in this specification, the term "reliability" may refer to the degree to which results appear consistently when data is repeatedly measured or analyzed under the same conditions. This serves as an important criterion for evaluating the reproducibility of results during the system's data collection and analysis process and can be considered essential to ensure data consistency and accuracy. In this invention, "reliability" is used to evaluate the repeatability of data collected from satellite-based positioning systems (GNSS) and inertial measurement units (IMU) and results derived during neural network analysis, and can serve as a key element in ensuring the stability and reliability of the system.

[0333] As used in this specification, the term "validity" may refer to the degree to which specific data or analysis results accurately reflect the target or standard intended for actual measurement. This can be used as a criterion to evaluate not only the reliability of the data but also whether the results are substantially appropriate and meaningful. In the present invention, "validity" is evaluated based on the correlation between existing standard data (gold standard), such as Global Navigation Satellite Systems (GNSS), Inertial Dead Reckoning, Inertial Measurement Units (IMU), and Motion Capture Systems (VICON), and can play an important role in ensuring the validity and appropriateness of the analysis results.

[0334] As used in this specification, the term "neural network analysis model" may refer to an artificial intelligence algorithm structure that processes input data to learn specific patterns and analyzes or predicts results based thereon. This may include one or more neural network structures selected from a group consisting of an Artificial Neural Network (ANN), a Multilayer Perceptron (MLP), a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), and a Long Short-Term Memory Network (LSTM), based on an Artificial Neural Network (ANN). The term "neural network analysis model" is used to analyze the spatial and temporal characteristics of data or to learn complex relationships, and the learned model may be designed to be applicable in various environments.

[0335] As used in this specification, the term "abnormal gait characteristics" refers to movement characteristics that deviate from the normal pattern observed during walking. These characteristics may manifest in various forms, such as reduced walking speed, limited range of motion, foot dragging, irregularity of walking rhythm, and abnormal knee flexion angles, and are detected by analyzing acceleration and angular velocity data collected through an inertial measurement device.

[0336] As used in this specification, the term "existing standard" may refer to reference values ​​or verified data referenced to evaluate the accuracy and validity of data. In the present invention, data collected from high-precision equipment such as a motion capture system (VICON) may be used as an existing standard, but is not limited thereto. In the present invention, data acquired from a motion capture system may be used as an existing standard to develop the algorithm of the reliability correction unit. The optimized algorithm data developed in this way may also become a new existing standard, and this may be utilized to verify reliability and validity in the position correction unit when a user wears an inertial measurement unit (IMU), but is not limited thereto. Furthermore, the present invention performs repeatability and concurrent validity evaluations based on this existing standard data to correct the data quality of the inertial measurement unit and increase reliability. Moreover, the existing standard is used as a standard for comparison and verification during the data analysis and neural network model training process, thereby enabling the continuous improvement of the performance of the optimized algorithm.

[0337] As used in this specification, the term “movement path data” refers to user movement trajectory information calculated from temporal changes in location information. It may include one or more data selected from the group consisting of movement distance, movement speed, direction change patterns, and movement characteristics by segment, and is not limited to a specific calculation method or map format.

[0338] As used in this specification, the term "degenerative brain disease" comprehensively refers to chronic and progressive diseases resulting from the gradual degeneration of the nervous system, and may include, for example, Parkinson's disease, Alzheimer's disease, Huntington's disease, and dementia. These diseases are characterized by the gradual deterioration of an individual's motor skills, cognitive functions, or ability to perform daily activities due to the decline and loss of neuronal function. "Degenerative brain disease" may manifest in various forms depending on specific symptoms, causes, or mechanisms of onset, and in this invention, it can be used as a central concept in the process of analyzing and processing data to detect and predict early signals of these diseases. "Degenerative brain disease" encompasses major disease categories addressed in early prediction and diagnosis utilizing neural network models, and may be interpreted to include various disease types depending on the technical purpose and scope of application.

[0339] As used in this specification, the term "essential tremor" refers to a type of neurological disorder characterized by unintended, regular, and rhythmic tremors that primarily manifest in the hands, head, and neck. It occurs without a specific cause and is often triggered by genetic factors or abnormal activity of the nervous system. "Essential tremor" may become more pronounced during states of tension or the performance of specific movements, and can cause discomfort in daily life. In the process of analyzing and diagnosing "essential tremor," the present invention focuses on evaluating the intensity, frequency, and pattern of the tremor by utilizing user movement data and an Inertial Measurement Unit (IMU). Through this, the degree of tremor can be identified, and appropriate exercise content or vibration feedback can be provided to contribute to symptom relief. In this invention, the term "essential tremor" may be used in various clinical settings and can be interpreted in diverse contexts, such as the alleviation of tremor symptoms, early detection, and preventive management.

[0340] As used in this specification, the term "bilateral lower limb contact" may refer to the action or state in which both lower limbs of a user simultaneously touch the ground while walking. This is an important process for ensuring stability and balance as part of the gait cycle, and it often occurs particularly when walking slowly or while stationary. "Bilateral lower limb contact" is frequently observed in patients with impaired balance or degenerative neurological diseases and can be used as an important indicator for evaluating gait stability or motor function. In this invention, to analyze the "bilateral lower limb contact" state, information collected from gait data and an inertial measurement device can be utilized to evaluate the user's balance and gait status in real time and provide feedback. Through this, exercise content suitable for the user can be provided, or abnormal gait patterns can be detected to contribute to the early prediction and prevention of degenerative brain diseases.

[0341] As used herein, the term “cognitive decline data” refers to data that quantifies one or more signs selected from a group consisting of declines in orientation, spatial perception, and judgment appearing in gait data and movement path data. It may be expressed in the form of relative indices or grades and is not limited to whether it directly corresponds to clinical cognitive tests.

[0342] The aforementioned cognitive impairment data and movement path data interact complementarily, playing a crucial role in distinguishing whether changes in a user's movement trajectory are merely deviations from the intended path or unintentional wandering caused by cognitive impairment. Since cognitive impairment data reflects the user's neurological state and movement path data represents actual behavioral patterns, the combination of the two datasets enables risk assessment with higher accuracy within the wandering detection module. This allows for the differentiation between simple wayfinding errors and wandering resulting from medical cognitive decline.

[0343] As used herein, the term “wandering” refers to unintentional movement characteristics involving repetitive or abnormal paths without an intended destination. It may include one or more forms selected from the group consisting of repetitive movement, disorientation, abnormal path deviation, and discrepancy with daily movement patterns.

[0344] As used herein, the term “reference movement pattern” refers to an individual’s normal movement pattern generated by accumulating and learning a user’s long-term movement paths and walking data. It may be represented by one or more models selected from a group consisting of statistical models, probabilistic models, and neural network-based models, and is not limited to a specific learning structure or data accumulation method.

[0345] As used herein, the term “inertial-based dead reckoning” refers to a method of estimating a relative trajectory by integrating acceleration and angular velocity data from an IMU when a position signal is interrupted or of degraded quality. It is not limited to drift correction or integration algorithm methods.

[0346] The satellite-based position information and inertial-based dead reckoning data work complementarily to contribute to the stable estimation of a user's location. Satellite-based position information provides absolute positional accuracy, but signals may be lost or quality degraded due to environmental factors. On the other hand, inertial-based dead reckoning continuously provides relative movement information regardless of signal interruption, but cumulative errors may increase over time. In this invention, through the complementary combination of these two data, it is possible to calculate a stable movement trajectory even during signal interruption, and the dead reckoning path can be precisely corrected when the satellite signal is re-received. This significantly improves the reliability of position information that serves as the basis for loitering judgment.

[0347] As used in this specification, the term “map matching” refers to a correction process that aligns an estimated movement trajectory with actual roads or indoor structures on a map. It may include one or more methods selected from the group consisting of nearest-to-nearness matching, candidate-based matching, probabilistic matching, and Kalman filter-based matching, and is not limited to a specific algorithmic method.

[0348] As used herein, the term “alarm, route guidance, and return guidance information” refers to notifications, route information, and / or return assistance information provided to a user or guardian when wandering is detected. It may include one or more forms selected from the group consisting of visual information, voice guidance, and vibration feedback, and is not limited to a specific output method or data format.

[0349]

[0350] Preferred embodiments according to the present invention will be described in detail below with reference to the accompanying drawings. In describing the present invention below, if it is determined that a detailed description of known functions or configurations may unnecessarily obscure the essence of the present invention, such detailed description will be omitted. Unless otherwise specifically defined, all terms in this specification have the same general meaning as understood by a person skilled in the art to which the present invention pertains. In the event of a conflict with the meaning of a term used in this specification, the definition used in this specification shall prevail.

[0351]

[0352] The loitering detection system based on the position measuring device and inertial measurement unit of the present invention utilizes a position measuring device (GNSS) and a low-cost inertial measurement unit (IMU) attached to the user's body to collect the user's body movements, gait data, and movement path data in real time, and based on this, calculates data on cognitive decline related to degenerative brain diseases, thereby providing a technology capable of reliably determining whether wandering is occurring. The present invention proposes a system capable of integrally performing a series of flows from the acquisition, analysis, judgment, and provision of feedback of such data, and simultaneously presents a loitering detection method or a method for providing loitering detection information that includes the processing steps performed by such a system.

[0353] More specifically, the wandering detection method or the wandering detection information provision method of the present invention may include a series of procedures comprising: (i) collecting user data through a location measuring device and an inertial measuring device; (ii) analyzing the collected data based on a neural network to calculate movement paths and walking characteristics; (iii) generating cognitive impairment data through walking analysis; (iv) integrating the movement path data and cognitive impairment data to determine whether wandering has occurred; and (v) providing necessary feedback to a user or guardian according to the determination result.

[0354] Meanwhile, the loitering detection method or the loitering detection information provision method of the present invention may be realized through a loitering detection system composed of hardware and software components for implementing the above steps, and the system may be composed of a data collection module, a neural network analysis module, a gait analysis module, a loitering determination module, and a feedback provision module. These components may be designed to operate individually or interact with each other to perform the loitering detection function of the present invention.

[0355] In order to clarify the technical implementation of the present invention, the specific configuration and operation process of the loitering detection system will be described in detail below.

[0356]

[0357] FIG. 1 is a schematic diagram of a position measuring device and inertial measuring device-based loitering detection system according to an embodiment of the present invention. The position measuring device and inertial measuring device-based loitering detection system may include a data collection module (100), a neural network analysis module (200), a gait analysis module (300), a loitering determination module (400), and a feedback provision module (500).

[0358] The data collection module (100) can perform the function of collecting location information in real time using a satellite-based position measuring device (GNSS) attached to the user's body, and measuring and collecting acceleration and angular velocity data in real time using a low-cost inertial measurement device (IMU). The position measuring device may be composed of one or more GNSS (Global Navigation Satellite System) based position measuring devices selected from the group consisting of GPS, GLONASS, Galileo, and BeiDou, but is not limited thereto and may be replaced with a GNSS receiver of various forms and performance or an equivalent position measuring device. In addition, the inertial measurement device may be composed including a 3-axis accelerometer and a 3-axis gyroscope sensor, but the inertial sensor used in the present invention is not limited thereto and may be replaced with a sensor of various forms and performance.

[0359] The above satellite-based positioning device (GNSS) can be attached to the user's body, preferably to one or more parts selected from the group consisting of shoes, ankles, waist, chest, or shoulders, to reliably measure changes in the user's absolute position and movement path. For example, if the GNSS receiver is attached to the shoes or ankle area, it can more accurately reflect the point of contact with the ground and even minute changes in the movement trajectory during walking, and if attached to the waist or upper body, it can efficiently and reliably detect changes in the user's overall direction of movement and speed. In the present invention, such attachment locations are not limited to these, and can be selectively attached to various locations depending on user convenience, data reliability, and purpose of use.

[0360] The above-mentioned inertial measurement unit (IMU) can also be attached to various parts of the body, preferably to one or more parts selected from the group consisting of shoes, ankles, waist, chest, or shoulders, to precisely measure changes in acceleration and angular velocity and real-time movement patterns occurring during the user's walking. In particular, when the inertial measurement unit is attached to both the upper and lower parts of the knee, changes in the knee joint's flexion and extension angles, joint rotation speed, and range of motion (ROM) occurring during walking can be measured with high precision, which makes a significant contribution to detecting the user's walking characteristics, balance ability, and signs of abnormal walking. Additionally, when the unit is attached to the upper extremities (vertebral vertebrae C7–T1, superior angle of the scapula, rib cage, etc.), it is possible to analyze inconsistencies in upper and lower body movements, reduced balance, and upper body swaying patterns in detail, thereby increasing the reliability and validity of the data in detecting abnormal mobility characteristic of wandering.

[0361] More preferably, the inertial measurement device may be attached to the upper and lower parts of the knee and simultaneously attached to one or more upper body parts selected from the group consisting of specific spinal locations (C7–T1), the superior angle of the scapula, and the rib cage. By attaching sensors to the upper body, the motor coordination and balance status between the upper and lower body can be evaluated more precisely. In particular, sensors attached to the shoulders can capture upper body sway and balance changes during walking more accurately than when sensors are attached only to the lower body. Attaching the inertial measurement device can increase the reliability and validity of cognitive decline data in detecting movement discrepancies or abnormal patterns between the upper and lower body at an early stage.

[0362] In addition, the attachment locations of the GNSS-based position measurement device and the inertial measurement unit (IMU) act complementarily to provide technical advantages that improve the quality and interpretation accuracy of movement path data and gait data. For example, when the GNSS device is attached to a location that is relatively stable and reflects the entire direction of movement of the human body, such as the upper body center (waist, chest, shoulders), absolute position changes and movement trends can be measured more accurately. On the other hand, by attaching the IMU to areas where gait and joint movements are distinct, such as the ankles, knees, and chest, it is possible to measure gait rhythms, joint flexion and extension patterns, ground impact timing, and upper-lower body coordination at high resolution, which are difficult to capture with GNSS alone.

[0363] As described above, by reflecting different biomechanical displacements at the attachment positions of the GNSS device and the IMU device, a synergistic effect occurs in which changes in absolute coordinates based on GNSS and relative motion characteristics based on IMU are temporally aligned and mutually corrected. Specifically, while the GNSS device stably tracks changes in the direction and velocity of upper body movement, the IMU precisely provides rotation, vibration, and impact data centered on the lower limb joints to form a reference frame for correction between the two devices. As a result, instantaneous positional jitter or deviations occurring in the GNSS signal can be eliminated or mitigated through comparison with IMU-based motion patterns, while, conversely, drift occurring in the IMU can be stabilized over the long term using GNSS absolute coordinates.

[0364] For example, while the upper-body attached GNSS device consistently calculates the user's forward direction and speed of movement, the GNSS position change is realigned based on the joint rotation speed and heel-strike provided by the knee and ankle attached IMUs, thereby maintaining the continuity of the movement path in minute discontinuous sections of the path or in indoor entry sections. Conversely, in outdoor environments where the GNSS signal is stable, the GNSS absolute position tracking result offsets IMU drift to provide a reference frame for gait cycle and ROM analysis, thus minimizing cumulative error even during long-term movement.

[0365] In this way, most preferably, the GNSS device is attached to at least one of the user's waist, chest, or shoulder, and the IMU can be attached to at least one of the user's ankle, upper knee, or lower knee.

[0366] The combination of attachment locations of the above-mentioned GNSS device and IMU sensor does not merely diversify the data collection sites, but acts as a key factor in determining the quality of data fusion between GNSS and IMU, and forms a technical foundation that significantly improves the accuracy of movement path estimation, the precision of gait analysis, and ultimately the reliability of wandering judgment.

[0367] Meanwhile, the GNSS and IMU mentioned above may be attached separately to the user's body parts as independent devices as described above, but they may also be configured in a form integrated into a single sensor module.

[0368] The integrated sensor module may include a structure in which an IMU comprising a GNSS chipset, a 3-axis accelerometer, and a 3-axis gyroscope is embedded together within a single housing, and in this case, a GNSS receiving antenna, a Micro Controller Unit (MCU), a wireless communication module (BLE, Wi-Fi, or LTE-M), a battery, and a power management circuit (PMIC) may be packaged as a single unit.

[0369] The above-described integrated sensor module acquires absolute position information and inertial-based motion information simultaneously from a single device, thereby reducing the number of sensors that need to be attached to the user and improving wearability. Furthermore, it has the advantage of significantly improving data alignment by ensuring hardware-based clock synchronization between the GNSS and IMU. More preferably, the timestamp of the GNSS signal and the sampling clock of the IMU are managed integrally within the same MCU, so that time errors, sample drift, and latency that may occur during the GNSS-IMU fusion process can be minimized.

[0370] In addition, since the integrated sensor module is attached as a single unit to a specific part of the user's body, the GNSS signal reception status and the IMU sensor output are based on the same physical reference frame, allowing for immediate correction of GNSS signal fluctuations in indoor, outdoor, and complex environments into IMU-based relative motion patterns. For example, when the integrated module is attached to the user's chest or waist area, the GNSS stably measures the forward direction of movement and changes in absolute position, while the IMU simultaneously provides upper body sway, rotational speed, and changes in vertical and horizontal acceleration due to walking, thereby enabling much more precise trajectory estimation than with GNSS alone or IMU alone.

[0371] This integrated sensor module can be inserted into or attached to various types of wearable devices, such as inside shoes, ankle straps, waist belts, chest straps, and shoulder wearable devices. Since it can secure both location and inertial information with just one module, it can be efficiently applied even in environments where it is difficult to wear multiple sensors, such as for the elderly, dementia patients, and residents of care facilities.

[0372]

[0373] As described above, since the GNSS and IMU are attached to each part of the user's body or configured as an integrated module to provide position information and inertial information in real time, the data collection module (100) of the present invention is configured to collect and sort raw signals transmitted from these sensors in chronological order and to simultaneously monitor the quality and reception status of each signal. In particular, since the absolute position information provided by the GNSS and the high-frequency sampling-based motion information provided by the IMU have different data characteristics, the data collection module performs the role of processing the two information into a real-time stream by combining them into a synchronized form.

[0374] A data collection module (100) according to one embodiment of the present invention can collect position information, acceleration, and angular velocity data by sampling at arbitrary intervals between 0.001 seconds and 1 hour. Preferably, instantaneous movements, minute changes in direction, and rapid changes in speed occurring during walking can be stably captured through a high-speed sampling period of 0.01 seconds or less, and time synchronization between movement path data and walking data can be maintained more precisely. Such precise sampling and real-time data streams enable the detection of loss of direction, repetitive paths, and abnormal stop-and-return patterns, which are critically analyzed in the wandering judgment algorithm. Meanwhile, if the sampling interval is set to less than 0.001 seconds, unnecessary load is generated on the system processing unit due to excessive data generation, and data quality may be degraded as GNSS signal reception noise and minute vibrations of the IMU are excessively reflected. Conversely, if the sampling interval becomes excessively long, such as more than one hour, it fails to properly reflect the user's actual movement flow, resulting in a disruption of the continuity of the movement path and significant limitations in detecting abnormal movement patterns characteristic of wandering, such as path deviation and repetitive movement. Therefore, in one embodiment of the present invention, an appropriate sampling period may be set considering the quality, reliability, real-time nature, and processing efficiency of the data.

[0375] GNSS signal reception strength can become unstable in structures such as indoors, urban canyons, or tunnels, which may result in instantaneous position jumps, signal drops, and increased deviation values. To solve these problems, the data acquisition module of the present invention evaluates the quality indicators of the GNSS signal and the variability of position information in real time. As a result of the evaluation, if the reliability of the GNSS signal drops below a set threshold or the amount of position change exceeds the normal movement range, the system automatically switches to an inertial-based dead reckoning mode to continuously maintain a relative movement trajectory by integrating the acceleration and angular velocity data of the IMU. At this time, although IMU-based dead reckoning does not provide an absolute standard for GNSS position information, it has the advantage of maintaining a continuous and relative movement trajectory even in signal interruption sections, thereby serving to compensate for GNSS interruptions. When the GNSS signal is received stably again, the present invention compares the accumulated error between the absolute position provided by the GNSS and the IMU-based estimated trajectory and automatically corrects it through map matching based on map information or a Kalman filter-based fusion technique. According to this complementary connection structure, the GNSS provides absolute position accuracy and the IMU ensures continuity in signal interruption sections, thereby forming a dual safety mechanism in which the two signals complement each other's limitations. Consequently, the GNSS-IMU fusion structure performs high-precision path calculation centered on the absolute position during normal GNSS signal sections, maintains path continuity using IMU dead reckoning during unstable GNSS sections, and immediately corrects the accumulated error of the IMU trajectory upon GNSS recovery, thereby enabling the reliability of the entire movement path data to be stably maintained throughout the system.

[0376] The basic axis alignment of an inertial measurement unit (IMU) may vary due to differences in the device attachment position, misalignment of the attachment direction, and the user's wearing habits, and drift may accumulate during prolonged use. To solve these problems, the neural network analysis module (200) according to the present invention may include a reliability correction unit for normalizing the quality of data and a position correction unit for correcting the sensor attachment state.

[0377] The reliability correction unit receives acceleration and angular velocity time series data collected from an inertial measurement unit (IMU) and performs a preprocessing process to ensure the reproducibility and consistency of the data. More specifically, the reliability correction unit may perform preprocessing including data interpolation, axis correction, outlier removal, feature scaling, and normalization. Through this, signal distortion caused by sensor bias, drift, environmental noise, etc., can be reduced, and the reliability and validity of the input data used in the gait analysis module (300) and the wandering judgment module (400) can be maintained at a certain level or higher.

[0378] Meanwhile, the position correction unit can perform the function of estimating the actual attachment position and direction of the inertial measurement unit (IMU), determining whether the attachment state deviates from a predefined standard range, and providing a direct sensor attachment guide to the user based on the result. More specifically, the position correction unit analyzes one or more feature values ​​selected from a group consisting of acceleration and angular velocity patterns that appear repeatedly during the gait cycle, the average direction of the gravity vector, and gait symmetry indicators to determine whether the position or direction of the sensor is inconsistent with the standard value, and can classify specific types of attachment abnormalities such as attachment position errors (e.g., attachment on the front / rear of the ankle), misalignment of the attachment angle, and sensor rotation.

[0379] The position correction unit described above can automatically generate a sensor adjustment guide that clearly presents the necessary adjustment direction to the user based on the analysis results. For example, it can provide specific guidance information for user actions, such as instructing the user to rotate the sensor to align it with a reference axis if the attachment angle error exceeds a specific threshold, or inducing re-attachment to a designated location if it is determined that the attachment position has deviated from the intended body part. The sensor adjustment guide in the present invention is automatically selected and displayed by the system in accordance with the output characteristics of the terminal, such as visual, auditory, and vibration, and the guide information generated within the position correction unit can be transmitted directly to the user terminal without passing through a separate module.

[0380] In this way, the position correction unit does not merely determine whether the sensor is misaligned, but also provides specific adjustment information in real time to restore the attachment status to a normal standard, thereby inducing user intervention with minimal procedures. Through this, data quality degradation caused by abnormal attachment of the inertial measurement unit (IMU) can be preemptively blocked, and the consistency and reliability of the input signals provided to the neural network analysis module (200) and gait analysis module (300) can be continuously maintained.

[0381] As described above, the refined and synchronized location information, acceleration, and angular velocity data are input into the neural network analysis module (200) to produce movement path data and walking data, respectively. The produced walking data is then transmitted to the walking analysis module (300) to quantitatively evaluate the degree of progression of degenerative brain disease based on key walking characteristics such as walking speed, walking cycle, and contact time of both lower limbs, and to generate cognitive decline data by detecting abnormal walking characteristics. This cognitive decline data is transmitted to the wandering judgment module (400) along with the movement path data, and the wandering judgment module comprehensively analyzes movement path deviation, loss of direction, breakdown of walking rhythm, and cognitive decline patterns to use as key evidence to determine whether the user's movement is intentional movement or unintentional wandering. That is, position and inertial data originating from the data collection module are refined into movement path / walking data through the neural network analysis module (200), interpreted into cognitive function deterioration data through the walking analysis module (300), and finally integratedly judged by the wandering judgment module (400), thereby configuring a stepwise structure that simultaneously secures precision and reliability of wandering judgment.

[0382]

[0383] A neural network analysis module (200) according to an embodiment of the present invention receives preprocessed position information, acceleration, and angular velocity data as described above, and fuses and interprets GNSS-based absolute position information and IMU-based relative movement information through a neural network structure, thereby producing high-dimensional feature data that can be directly utilized for determining wandering. More specifically, the neural network analysis module (200) receives time-synchronized position and inertial time series data in the form of a window unit or a continuous time series, and applies one or more neural network analysis models selected from a group consisting of a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), a Long Short-Term Memory Network (LSTM), and a Transformer, thereby learning non-linear relationships inherent in movement path patterns and walking patterns. Through this, it is possible to generate feature vectors that simultaneously reflect movement characteristics unique to wandering, such as disorientation, repetitive paths, and abnormal stopping and regressing patterns, which are difficult to capture with simple speed and distance-based analysis, as well as walking abnormality characteristics such as irregular walking rhythm and upper and lower body incoordination.

[0384] FIG. 2 is a diagram showing the configuration of a neural network analysis module (200) according to one embodiment of the present invention.

[0385] The above neural network analysis module (200) may include a movement path calculation unit (210) and a walking data calculation unit (220).

[0386] The above movement path calculation unit (210) can perform the function of calculating movement path data by receiving location information data obtained through the data collection module (100) and analyzing the user's movement direction, movement distance, instantaneous speed, rotation angle, and path change rate.

[0387] Preferably, GNSS-based absolute position coordinates can be aligned in a time series form, and then various signal processing techniques such as a Kalman filter, a particle filter, or a moving average-based filter can be applied to minimize position noise and jitter. In addition, the movement path calculation unit (210) can calculate not only simple position coordinates but also high-dimensional features useful for detecting loitering characteristics, such as velocity profiles, direction vectors, and rates of change between movement segments.

[0388] In addition, the movement path calculation unit (210) includes one or more neural network structures selected from a group consisting of a multilayer perceptron (MLP), a convolutional neural network (CNN), a recurrent neural network (RNN), and a long short-term memory network (LSTM), based on an artificial neural network (ANN), and can simultaneously receive GNSS absolute coordinates and inertial-based dead reckoning data to calculate a movement path vector that reflects the complementary characteristics of the two data. Specifically, when the GNSS signal is partially lost, it prioritizes reflecting the IMU-based movement estimation path, and when the GNSS signal is recovered, it can automatically generate a corrected path. This neural network-based path fusion technique contributes to more stably estimating the user's actual movement trend.

[0389] The above gait data calculation unit (220) can perform the function of calculating gait data by analyzing the user's gait pattern based on acceleration and angular velocity data input from an inertial measurement unit (IMU). Preferably, the gait data calculation unit (220) can divide the preprocessed IMU time series signal into window units and automatically learn and calculate gait characteristic values ​​using a CNN-based feature extractor or an LSTM-based time series model. For example, the gait data calculation unit (220) can calculate various gait characteristic values ​​including cadence, step length, gait cycle, ground contact time (GCT), plantar pattern change, and balance index, and can evaluate the coordination of the upper and lower limbs by analyzing the rotation angle and ROM data of the knee and ankle sensors together.

[0390] In addition, the walking data calculation unit (220) can automatically detect abnormal changes in the user's walking pattern, such as sudden increase in irregularity, foot dragging signals, asymmetrical walking, and misalignment of upper and lower body joint rotation, and this can be used as an important criterion for cognitive function deterioration data and judgment of the possibility of wandering in the wandering judgment module (400) thereafter.

[0391] The above-described gait analysis module (300) receives gait data generated through the neural network analysis module (200), comprehensively evaluates the user's gait stability, balance, and normality, and performs the function of generating cognitive function decline data based on the results. The above-described gait analysis module (300) may be configured to quantitatively measure subtle abnormalities in gait patterns by considering not only the characteristics of a single gait event but also a certain time interval or long-term change trend, thereby selectively capturing patterns that appear specifically in cognitive function decline rather than general changes such as a simple decrease in walking speed.

[0392] More specifically, the gait analysis module (300) can analyze various gait indicators, such as the rate of change in stride length, cadence variability, gait cycle asymmetry, instability of joint range of motion (ROM), foot drag signal, dynamic balance indicator, and the degree of desynchronization of lower limb sway relative to upper body sway. In addition, the present invention can calculate a deviation index based on short-term and long-term gait variation patterns and quantify the degree of deviation from a normal gait pattern to convert it into a cognitive gait deterioration index. The cognitive gait deterioration index is calculated by reflecting characteristics that appear in cognitive and motor function deterioration, such as irregularity of gait rhythm patterns, delay when changing direction, repetitive hesitation, and breakdown of dynamic balance, and can be used as a unique indicator that is distinguished from simple gait performance deterioration.

[0393] In addition, the gait analysis module (300) can classify the level of cognitive decline into three stages of 'high', 'medium', and 'low' by comparing the calculated cognitive decline index with a preset reference value or a user-customized learning reference value. For example, it can be classified as 'low' if the deviation index compared to the normal group is below a certain range, 'medium' if moderate fluctuations and irregularities are confirmed, and 'high' if there is a distinct breakdown of the gait cycle or upper-lower body coordination. This graded cognitive decline index is then directly transmitted to the wandering judgment module (400) and combined with movement path data, and can be used as a key judgment criterion to determine whether the user's movement pattern is an intended movement or an abnormal wandering pattern.

[0394] For example, in the early stages of Alzheimer's dementia, delays in turning direction (hesitation), subtle irregularities in walking rhythm, and a decrease in cadence may be observed, while in patients with Parkinson's disease, desynchronization of upper and lower body movements, reduced stride length, and reduced speed may be characteristic. The gait analysis module (300) is designed to reflect these clinical characteristics and can stably calculate the cognitive decline index by automatically extracting the corresponding features from IMU-based time-series data. The cognitive decline index calculated in this way allows for a more precise assessment of the risk level by reflecting the long-term change trends of each user and is utilized as information to increase the reliability of wandering judgment by the wandering judgment module (400).

[0395] The above-mentioned wandering judgment module (400) is a core module that ultimately determines whether wandering is occurring within the overall judgment structure of the present invention. It performs the function of evaluating the user's current movement status with high precision by integrating movement path data and cognitive decline data calculated through the neural network analysis module (200) and the gait analysis module (300). In particular, by simultaneously considering spatial deviations based on the movement path and neurological changes based on gait and cognitive status, it structurally compensates for the false positive and non-detection problems inherent in simple location-based analysis methods and can provide judgment results that reflect actual wandering risk. The wandering judgment module (400) of the present invention realizes wandering detection with much higher accuracy compared to existing systems through an integrated judgment system that reflects not only spatiotemporal abnormalities in movement path patterns but also complex degenerative signals such as changes in the user's cognitive function, decline in dynamic balance, and loss of sense of direction.

[0396] In addition, the wandering judgment module (400) can quantitatively determine whether the movement is an intended normal movement or an unintentional wandering caused by cognitive decline by utilizing a user-specific standard movement pattern generated based on the user's normal movement pattern and long-term change trend to compare and analyze real-time movement data. The wandering judgment module (400) of the present invention has a multimodal judgment structure that comprehensively considers three elements: GNSS-based location information, IMU-based gait and posture information, and a cognitive decline index. Through this, it is designed so that even with the same path deviation, the risk level is evaluated differently depending on the level of cognitive decline of the user. Furthermore, the wandering judgment result influences the update of the user-specific standard movement pattern, forming a cyclic learning structure that proceeds in the order of standard pattern generation, real-time judgment, and standard pattern correction, thereby evolving into a high-precision wandering detection system that reflects individual characteristics.

[0397] FIG. 11 shows a configuration in which a reference movement pattern generation unit (420) is added to a wandering determination module (400) according to an embodiment of the present invention. The wandering determination module (400) may further include a reference movement pattern generation unit (420) that generates a reference movement pattern for each user based on the user's walking and movement path patterns by accumulating and learning the wandering determination unit (410), the movement path data, and the cognitive function deterioration data.

[0398] More specifically, the above-mentioned wandering determination module includes a wandering determination unit that calculates an integrated risk level through the deviation index of the movement path data and the cognitive impairment data, wherein the wandering determination unit assigns different weights to the movement path data deviation index according to the level of the cognitive impairment data, and can determine the user's movement as unintentional wandering if the integrated risk level exceeds a preset threshold value.

[0399] The above-mentioned wandering determination unit (410) can perform the function of determining whether the user's current movement is intentional movement or unintentional wandering by comparing real-time input movement path data and cognitive impairment data with a standard movement pattern for each user.

[0400] The movement path data of the present invention includes a trajectory deviation index calculated by integrating GNSS-based absolute position changes and IMU-based gait and posture changes, which quantitatively indicates the degree to which the user's movement pattern deviates from a reference movement pattern. Meanwhile, the cognitive function degradation data is a neuro-motor degradation index calculated based on neurological characteristics such as gait rhythm irregularity, gait cycle asymmetry, hesitation when changing direction, stride reduction rate, and upper-lower body incoordination, which are calculated by the gait analysis module (300). This is a key element that reflects the degree of progression of degenerative brain diseases, such as a decline in spatial perception, sense of direction, and judgment ability, and even if the same movement path deviation is observed, the higher the level of cognitive degradation, the higher the risk of wandering can be interpreted.

[0401] A wandering determination unit (410) according to one embodiment of the present invention can significantly increase the reliability of the determination by not determining whether wandering is occurring solely based on abnormal signs based on the movement path, but by integrating cognitive function decline data calculated through the gait analysis module (300). Since the cognitive function decline data reflects a neurological state directly linked to a decline in sense of direction, spatial perception, and decision-making ability, the possibility of wandering can be interpreted differently depending on the level of cognitive function decline, even with the same movement deviation.

[0402] In addition, the wandering judgment unit of the present invention utilizes the cognitive function decline index input from the gait analysis module (300) as is, but may be configured to assign different risk weights even for the same movement path deviation depending on which stage the cognitive function decline index corresponds to among 'high', 'medium', and 'low'. For example, if classified as 'low' by the gait analysis module (300), the possibility of normal movement with minimal impact from cognitive function decline is evaluated highly; if classified as 'medium', sensitivity to abnormalities in the movement path is increased; and if classified as 'high', neurological problems such as reduced spatial perception or loss of sense of direction are considered to have a high probability of directly affecting the movement pattern, so a higher weight may be applied when calculating the wandering risk. Therefore, the wandering judgment unit (410) of the present invention can perform a much more precise and personalized wandering risk assessment than the conventional method based on simple movement deviation by combining and interpreting the cognitive function decline index pre-classified through the gait analysis module with movement path data.

[0403] That is, the wandering judgment unit (410) of the present invention includes a dynamic weight-based risk calculation structure that applies different weights to the movement path deviation index according to the level of the cognitive function deterioration data (e.g., high, medium, low), thereby enabling the distinction between normal movement caused by temporary situational factors and unintentional wandering caused by cognitive deterioration, and structurally compensating for the false detection problem of a simple location-based system.

[0404] Here, the cognitive decline data is data that quantifies the degree of cognitive decline of the user based on one or more indicators selected from a group consisting of gait rhythm irregularity, increased hesitation when changing direction, gait cycle asymmetry, stride length reduction rate, and upper-lower body incoordination, calculated through the gait analysis module (300). Unlike simple gait abnormality detection, this cognitive decline data reflects a neurological state directly linked to the decline in the user's spatial perception, sense of direction, and judgment ability; therefore, when combined with movement path data, it can serve as a key input for determining the accuracy of wandering judgment.

[0405] For example, even when signs of wandering based on movement paths are detected, such as abnormal path deviation, repetitive movement, or direction changes without a destination, if the user has a low degree of cognitive decline, it may be a temporary change due to situational factors; however, if the same movement pattern occurs when the cognitive decline index has risen above a certain level, it can be determined that there is a very high probability of unintentional wandering. Therefore, the wandering judgment unit (410) calculates the probability of wandering by comprehensively considering the spatial change (trajectory deviation) of the movement path data and the neurological change (neuro-motor degradation) of the cognitive decline data, and thereby achieves wandering detection with much higher accuracy than a simple location-based judgment method.

[0406] In addition, the above cognitive impairment data is reflected as a weighting factor in the process in which the wandering judgment unit (410) combines abnormal signs based on movement path and abnormal signs based on perception to make a judgment, and is designed so that different weights are applied to the risk assessment result depending on the level of cognitive impairment even if the movement deviation is the same. Through this, the wandering judgment algorithm of the present invention significantly reduces the risk of false positives and non-detections compared to the conventional simple speed and distance-based method, and can detect high-risk patterns combining cognitive impairment and movement abnormalities more sensitively.

[0407] The above-mentioned wandering judgment unit (410) utilizes one or more neural network structures selected from a group consisting of a multilayer perceptron (MLP), a convolutional neural network (CNN), a recurrent neural network (RNN), and a long short-term memory network (LSTM), based on an artificial neural network (ANN), to learn changes in movement path patterns and fluctuations in cognitive function decline indices as an intercorrelation structure, and automatically detects complex abnormal movements that deviate from normal movement patterns. This can provide higher accuracy than existing simple speed analysis or GNSS-based location comparison methods.

[0408] Unlike conventional technology that determines wandering based solely on movement path data, the above-mentioned wandering determination unit (410) can perform an integrated analysis of movement path-walking-cognition states by reflecting inertial measurement unit (IMU)-based walking characteristics, such as the coordination pattern of upper and lower body joints, the asymmetry of walking rhythm, and the correlation between upper body sway and lower body rotation patterns. This analysis reflects the inherent co-dependency between GNSS-based position information and IMU-based walking information in the reference movement pattern, thereby having a clear technical distinction from conventional technology that performed simple parallel analysis.

[0409] Meanwhile, the above-mentioned loitering judgment unit (410) can analyze the deviation between real-time movement data and a reference movement pattern to calculate not only the loitering judgment result but also the confidence score of the judgment, and the confidence information is provided to the reference movement pattern generation unit (420) so that it can be reflected in the weight adjustment when updating the model.

[0410] In this way, the wandering judgment unit (410) calculates an integrated risk score by reflecting the weights for the movement path deviation index and the cognitive function decline index. Additionally, if the integrated risk score exceeds a threshold set by the system, the movement is determined to be unintentional wandering. This threshold is set by reflecting the normal movement range, long-term movement characteristics, and environmental changes for each user, and can be updated as necessary.

[0411] Meanwhile, the reference movement pattern generation unit (420) according to one embodiment of the present invention defines a normal movement range based on the user's long-term movement characteristics and, by reflecting the long-term trend of cognitive function decline indicators such as walking stability, rhythm, and sense of direction, can generate a “normal movement pattern reflecting cognitive state” rather than a simple movement pattern. As a result, the reference movement pattern becomes not a simple spatial path module, but an advanced baseline that integrates and reflects the user's neurological and gait state.

[0412] The reference movement pattern generation unit (420) can update the reference movement pattern by reflecting changes in the user's life, such as new frequently visited places or changes in walking ability, over time, and the wandering judgment unit (410) can apply this updated pattern to perform a judgment based on the latest state. Through this interaction, the accuracy and personalization of the wandering judgment are improved.

[0413] The reference movement pattern generation unit (420) above does not merely generate a reference pattern using long-term accumulated data, but also receives the judgment result calculated by the wandering judgment unit (410) as input to dynamically correct or update the reliability of the existing reference movement pattern. That is, movement data that is repeatedly judged as normal movement over a certain period increases the weight of the reference movement pattern, while sections judged as unintentional wandering are considered as noise in the reference movement pattern and excluded or have their weight reduced. This bidirectional update structure is a core technical feature of the present invention that distinguishes it from existing static reference pattern-based systems.

[0414] As described above, when a structure is provided in which the reference movement pattern generation unit (420) learns the normal movement characteristics of each user over a long period to form a baseline and the wandering judgment unit (410) dynamically updates the reference pattern by reflecting the real-time judgment results, the wandering judgment unit (410) can more precisely analyze the shape and distribution of deviations in which real-time movement data deviates from the updated reference pattern. Thus, the process of generating the reference movement pattern and the process of wandering judgment are organically linked to each other, and unlike existing technology based on simple comparison, a cyclic judgment structure is completed that repeats in the order of reference pattern, judgment, reference pattern correction, and more precise judgment.

[0415] In addition, the loitering judgment unit (410) does not use the normal movement pattern generated by the reference movement pattern generation unit (420) as a simple comparison standard, but rather learns non-linear characteristics such as the distribution, fluctuation range, and temporal rate of change of the normal pattern to quantitatively calculate the degree to which real-time movement data deviates from the statistical boundary of the normal pattern. This deviation-based judgment structure is fundamentally different from the simple speed change or distance-based threshold determination used in existing GNSS-based loitering detection technology, and enables high-precision loitering judgment that reflects individual walking habits.

[0416] In conclusion, the wandering determination module of the present invention calculates an integrated risk level by applying different weights to the deviation index of the movement path data and the deterioration index of the cognitive function deterioration data, and if the integrated risk level exceeds a preset threshold, determines the user's current movement as unintentional wandering, or generates a determination result regarding whether the user's current movement is intentional or unintentional wandering by comparing the movement path data and the cognitive function deterioration data with the user-specific standard movement pattern, thereby enabling the determination of whether wandering is occurring through mutually complementary methods and allowing for the derivation of more reliable results.

[0417]

[0418] The feedback providing module (500) can perform the function of preventing dangerous situations early and inducing safe movement by providing immediate feedback, including an alarm, route guidance, and / or return guidance information, to the user terminal or guardian terminal when the user's movement is determined to be unintentional wandering by the wandering judgment module (400). The feedback providing module (500) can provide one or more alarm signals selected from a group consisting of vibration, voice, and visual notifications, and can be configured to automatically adjust the alarm intensity, duration, and repetition cycle according to the wandering risk level or cognitive function deterioration index. Through this, a dynamic alarm system that adapts to the user's state is implemented, rather than a simple fixed alarm method.

[0419] In addition, the feedback providing module (500) may include a map-based feedback function linked with GNSS / IMU-based movement path data. For example, it can visualize and display the direction of deviation from the current location, the distance of deviation, a safe return path, and the shortest path to a residence or a registered guardian location, and can provide various forms of feedback tailored to the user's condition, such as direction arrows, voice navigation, and high-contrast color guidance. This route guidance function is not limited to providing simple warnings but has technical differentiation by providing a continuous safety guidance process in the order of wandering detection, creation of a safe return path, and step-by-step guidance.

[0420] The above feedback providing module (500) can be linked in real-time with not only the user terminal but also the guardian terminal, and wandering-related information, such as real-time location information, records of the time, location, and path of wandering occurrence, recent trends in cognitive function decline index changes, and alarm occurrence history, can be automatically provided to the guardian terminal. The guardian terminal can perform step-by-step response functions, such as push notifications, emergency call connection, and automatic notification to a designated protection agency, depending on the risk level, thereby enabling active response based on the risk level of wandering, unlike a simple location sharing system.

[0421] In addition, it may include a preemptive feedback function that provides a prediction-based alert before the judgment of the loitering judgment module (400) when a user approaches a danger zone or stops abnormally for a long time at a specific coordinate. This configuration provides an alert by detecting a dangerous situation in advance based on a sudden change in movement patterns or environmental risk, and has a significant technical difference from conventional simple area-based geofencing technology in that it combines real-time danger zone monitoring and prediction functions.

[0422] Additionally, the feedback providing module (500) may include a learning function that gradually optimizes feedback sensitivity and alarm methods based on long-term changes in user-specific behavior patterns, movement speed, and cognitive decline index. For example, for a user with a hearing impairment, visual and vibration feedback may be reinforced first, or for a user whose cognitive decline is progressing rapidly, alarms may be automatically adjusted to be provided even at lower deviations. This user-customized feedback optimization function has greater technical advancement than existing uniform alarm methods and can be utilized as a practical differentiating point when registering a patent.

[0423] FIG. 12 shows a configuration in which a path correction unit (230) is added to the neural network analysis module (200) of FIG. 10.

[0424] The neural network analysis module (200) of the present invention may further include a path correction unit (230) to maintain the continuity and accuracy of the movement path data calculated by the movement path calculation unit (210). The path correction unit (230) performs the function of automatically detecting and correcting movement path discontinuities, instantaneous position jumps (jitter), and increases in cumulative errors that occur in signal interruptions or quality degradation sections of the satellite-based position measurement device (GNSS).

[0425] Specifically, the path correction unit (230) evaluates the reliability of the GNSS signal by utilizing one or more criteria selected from a group consisting of GNSS signal status information, the rate of change of position, and the results of verifying the consistency of velocity and acceleration. As a result of the evaluation, if the reliability of the GNSS signal drops below a set reference value or the amount of change in position deviates from the normal movement pattern, the path correction unit (230) may be configured to automatically switch to an IMU-based dead reckoning mode and maintain a relative movement trajectory by integrating acceleration and angular velocity data.

[0426] The above inertial-based dead reckoning navigation enables the maintenance of a continuous trajectory even in the absence of GNSS signals, and IMU drift accumulated over time is corrected by comparing and matching with accurate absolute position coordinates at the point when GNSS signals are restored.

[0427] When the GNSS signal is successfully re-received, the path correction unit (230) calculates the error between the GNSS-based absolute position coordinates and the IMU-based relative movement trajectory, and corrects it using one or more methods selected from a group consisting of map information-based map matching, Kalman filter-based fusion (EKF·UKF), factor graph-based optimization (Graph-SLAM), and deep learning-based trajectory refinement models.

[0428] This correction process enables the continuity and accuracy of the movement path to be maintained even in GNSS-vulnerable environments such as indoors, urban canyons, and tunnels. Additionally, the path correction unit (230) detects abnormal coordinate values, instantaneous position spikes, and non-physical changes in velocity and acceleration that may occur during the GNSS-IMU fusion process, and minimizes distortion of the movement path data by removing or interpolating the corresponding sections. Through this, the stability and reliability of the movement path pattern used for real-time wandering judgment can be secured.

[0429] Accordingly, the path correction unit (230) is a key component that ensures the accuracy and continuity of the entire movement path data by aligning the GNSS-based absolute position signal and the IMU-based relative movement estimate in time, maintaining the IMU-based estimated trajectory in the GNSS signal interruption section, and precisely correcting the accumulated error through map-based map matching or filter-based fusion when the GNSS signal is restored.

[0430]

[0431] Meanwhile, the wandering detection system according to the present invention may be composed of a series of AI-based processing pipelines that use raw data collected through a GNSS-based position measuring device and an inertial measuring device (IMU) as input values ​​to calculate movement path deviation indicators, walking characteristic indicators, and cognitive function deterioration indicators in stages, and integrate and input these into a neural network-based AI learning model to finally automatically determine whether wandering is occurring.

[0432] First, the data collection module acquires time-synchronized GNSS position data and IMU inertial data in real time. Through a preprocessing process, the raw data undergoes GNSS drift correction, IMU noise removal, and time-axis alignment between the data. Subsequently, the path shape, rate of change of direction, and trajectory deviation are calculated through the trajectory calculation unit. At the same time, the gait data calculation unit extracts gait characteristics such as gait rhythm, stride length change, left-right asymmetry, and upper-lower body coordination from the inertial signal, and the gait analysis module calculates the level of cognitive decline, including 'high, medium, and low' stages, based on these characteristics.

[0433] The preprocessed GNSS-based movement path indicators and IMU-based gait and perception indicators are combined again as input values ​​for the AI ​​learning module. The AI ​​learning module includes a structure selected from a Multilayer Perceptron (MLP), Recurrent Neural Network (RNN), Long Short-Term Memory Network (LSTM), or Transformer-based model, and can learn the correlation between changes in movement paths and changes in gait characteristics as an internal model parameter through pre-training on normal movement patterns and actual wandering patterns. During the learning process, the level of cognitive decline is reflected as an adjustment factor for the weight parameter applied to the movement path deviation indicator, and the model automatically optimizes the parameter so that the same movement path deviation is interpreted as a higher risk level as cognitive decline becomes more severe.

[0434] The trained AI model calculates an integrated risk score by integrating movement path deviation indicators and cognitive decline indicators input in real time. The integrated risk score is expressed as a probability value in the range of 0 to 1 or a normalized risk value, and if it exceeds a reference threshold set in the present invention, the movement is determined to be unintentional wandering. Unlike simple speed change or distance-based comparison methods, this determination precisely reflects changes in temporal patterns, upper and lower body coordination patterns, and the influence of cognitive decline on the movement path; thus, it provides significantly higher accuracy and adaptability compared to existing rule-based wandering detection methods.

[0435] Furthermore, the reference movement pattern generation unit automatically adjusts the baseline according to changes in the user's lifestyle by continuously updating normal movement patterns based on the judgment results of the AI ​​model and long-term movement data, and the loitering judgment unit enables more dynamic and personalized judgment by recalculating the integrated risk level of subsequent data based on the latest reference pattern. This iterative update process forms a cyclical structure in the form of reference pattern generation, real-time judgment, judgment result feedback, and reference pattern correction, which constitutes the core technical differentiation of the AI-based loitering detection system provided by the present invention.

[0436] In addition, the wandering detection system according to the present invention may further include a configuration that maintains the wandering detection function even when the smartphone is not carried or the BLE connection is disconnected by constantly monitoring the short-range wireless connection status between the user's wearable terminal and the user terminal (e.g., smartphone) via BLE (Bluetooth Low Energy).

[0437] Specifically, the wearable terminal can be attached to any body part suitable for movement analysis, such as the user's ankle, knee, waist, or chest, and includes local feedback means such as a BLE communication module, an LTE communication module, a vibration motor, and / or an LED, in addition to GNSS and IMU sensors. The control unit of the wearable terminal periodically checks the BLE connection status and recognizes it as an abnormal state if the connection is disconnected, and can intuitively provide the user that "the connection with the smartphone has been disconnected" by first outputting a vibration or lighting an LED.

[0438] If the BLE connection is not resumed within a predetermined time (e.g., tens of seconds to several minutes) despite such a first notification, the on-device event detection unit within the wearable terminal can perform a simplified evaluation of candidate wandering events using location and walking data directly collected from GNSS and IMU. At this time, key indicators such as movement path deviation, signs of disorientation, and abnormal stopping and returning patterns are calculated, and the wearable terminal is configured to calculate the wandering risk level itself based on the results.

[0439] If the loitering risk calculated by the on-device event detection unit exceeds a preset threshold, the wearable terminal transmits a loitering alert directly to a server or guardian terminal using an LTE communication module instead of communication via a smartphone using BLE. Subsequently, a feedback provision module linked with the server or guardian terminal provides immediate follow-up feedback, such as push notifications, location information display, and phone connection, thereby realizing a dual safety mechanism that ensures the loitering detection and alert functions are not interrupted even in situations such as failure to carry a smartphone, battery depletion, or BLE communication failure.

[0440] Furthermore, the wandering detection system according to the present invention can be configured to quantitatively evaluate the long-term risk level for each user by integrally analyzing cognitive decline data calculated by the gait analysis module and the wandering risk level calculated by the wandering judgment module on the server side. The cognitive decline data reflects abnormal gait signs that appear prior to the cognitive decline stage, such as instability of gait rhythm, left-right asymmetry, stride length variation coefficient, and upper-lower body coordination patterns, and these indicators exhibit characteristics of having a progressive trend over time. Meanwhile, the wandering risk level may consist of indicators sensitive to short-term changes, such as deviation from the actual movement path, patterns of disorientation, and the rate of increase in distance to the destination. In the present invention, by combining these short-term risk indicators (wandering risk level) and long-term state indicators (cognitive decline data) on the server side, it is possible to more precisely identify at what speed and direction the user's movement and gait functions are changing.

[0441] To this end, the server may include an integrated risk calculation unit that calculates a single integrated risk value by combining the level of cognitive decline and the wandering risk. When calculating the integrated risk, weights may be dynamically applied so that if cognitive decline exceeds a certain level, the same wandering risk is interpreted as a higher long-term risk; conversely, if the level of cognitive decline is low, individual differences may be reflected by weighting the abnormal change in the wandering risk indicator. Furthermore, the server can generate or update personalized baselines based on long-term accumulated user movement data, gait characteristic statistics, and changes in movement patterns by season and time of day, and periodically recalculate the integrated risk by reflecting the deviation from these baselines. Unlike a simple cumulative average method, this structure is designed to consider temporal trend analysis, persistence measures, and volatility indicators together, thereby enabling the detailed capture of both rapid increases in risk and gradual trends in cognitive decline.

[0442] The aforementioned integrated risk level can be utilized for active intervention functions aimed at mitigating the decline in a user's cognitive and gait functions or reversing the progression of risk, going beyond simple monitoring. Specifically, the server may select at least one electronic medicine-based intervention protocol, consisting of dual-task training for gait perception, balance function enhancement training, and lifestyle adjustment tasks (such as sleep, dietary, and activity level adjustments), by comprehensively analyzing the absolute value of the integrated risk level, the magnitude of the increase compared to the past, and the presence of rapid variability. In this case, the selection of the electronic medicine protocol can be implemented not merely by presenting tasks predetermined based on risk level ranges, but by analyzing which gait elements most strongly reflect cognitive decline in a specific user and automatically recommending intervention elements optimized for that user.

[0443] The selected electronic drug protocol is transmitted to a mobile or wearable terminal, and by performing detailed training procedures or lifestyle adjustment tasks guided by said terminal, the user can delay the long-term progression of cognitive decline or prevent abnormal movement patterns from developing into actual wandering. For example, if a specific user exhibits a pattern of rapidly deteriorating gait balance during cognitive dual-task performance, the system may prioritize dual-task-based gait training; conversely, if gait variability is low but cognitive reaction speed continues to decline, lifestyle adjustment tasks may be prioritized.

[0444] Meanwhile, the wandering detection system according to the present invention may further include a fall detection unit that performs an IMU-based fall detection function in the neural network analysis module. The fall detection unit is configured to detect specific sequences associated with a fall by taking acceleration and angular velocity data acquired in real time from an IMU sensor embedded in a wearable terminal as input. For example, a fall event generally has the characteristic of continuously appearing patterns such as a low-acceleration signal appearing during the free-fall phase, an impact acceleration peak at the time of impact with the ground, an inactivity phase after the impact, and an abnormal orientation shift occurring after the fall. The fall detection unit can detect a candidate fall event by analyzing at least two of these patterns.

[0445] Additionally, the fall detection unit may be configured to include either a neural network-based time series analysis module (e.g., MLP, RNN, CNN, or Transformer) or a rule-based algorithm to calculate a confidence score for a detected candidate event. If the confidence score is greater than or equal to a preset threshold, the wearable terminal may generate an event message containing a fall flag and GNSS location information and immediately transmit it to a server or guardian terminal via an LTE communication unit, regardless of whether the BLE connection status is normal. In this case, the fall detection unit may operate independently of the on-device event detection unit or be implemented as a sub-function within the module, so that the fall detection function can be reliably performed in parallel with the wandering detection function.

[0446] Furthermore, the aforementioned fall event message transmitted to the server or guardian terminal can be linked with a feedback provision module to automatically execute follow-up response procedures. For example, the server can synthesize the reliability of the fall flag, location information, and past fall history to immediately provide a push notification to the guardian terminal, and if necessary, resend voice, vibration, or LED alerts to the wearable terminal to guide the user to request help themselves if they are conscious. Additionally, situation response functions such as "phone connection," "linking to 119 emergency call," "real-time location tracking," and "voice call request" can be immediately activated through the guardian terminal interface.

[0447]

[0448] As described above, the satellite-based positioning device and inertial measurement device-based wandering detection system and the wandering detection method using the same according to the present invention can precisely collect various biosignals and movement information obtained through the positioning device and inertial measurement device attached to the user's body, calculate movement path data and gait data through neural network-based analysis, and determine whether wandering is occurring with high reliability by integrating this with cognitive decline data calculated through gait analysis. In addition, by including various correction and supplementation functions such as inertial-based dead reckoning navigation to compensate for GNSS signal interruption and deviation, a position correction unit to correct the sensor attachment status, and a reliability correction unit to normalize data quality, it can provide stable and consistent wandering detection results regardless of the user's movement environment or wearing method.

[0449] Furthermore, the system of the present invention enables accident prevention and real-time response by providing real-time alerts, route guidance, and return guidance information to a user terminal or guardian terminal based on the result of wandering judgment, and can implement a customized wandering detection function that reflects individual long-term movement and walking characteristics through a standard movement pattern generation function.

[0450]

[0451] The embodiments of the present invention have been described above. The technical features are not limited thereto and include all variations that can be modified and applied by a person skilled in the art based on the technical concept of the present invention.

Claims

1. A data collection module that measures acceleration and angular velocity data of an inertial measurement device attached to a user's body part in real time and collects the measured data; A neural network analysis module that calculates walking data by analyzing data collected through the above data collection module using a neural network analysis model; A gait analysis module that analyzes the above gait data to detect the degree of progression of degenerative brain disease or abnormal gait characteristics; and An inertial measurement device-based degenerative brain disease prevention system characterized by including a feedback providing module that provides real-time feedback upon detection of the above-mentioned abnormal gait characteristics.

2. In Paragraph 1, An inertial measurement device-based degenerative brain disease prevention system characterized by the above-mentioned inertial measurement device being attached to one or more body parts selected from the group consisting of the chest, shoulders, back, knees, and feet.

3. In Paragraph 1, The above neural network analysis module is, An inertial measurement device-based degenerative brain disease prevention system comprising a reliability correction unit that improves the consistency and accuracy of data collected through the above-mentioned data collection module, wherein the reliability correction unit performs preprocessing including data interpolation, axis correction, outlier removal, feature scaling, and normalization processes.

4. In Paragraph 1, The above neural network analysis module is, An inertial measurement device-based degenerative brain disease prevention system characterized by including a position correction unit that analyzes data collected through the above-mentioned data collection module and provides a guide to the user to adjust the sensor attachment position.

5. A content delivery module that provides customized content tailored to the user's state; A data collection module that measures acceleration and angular velocity data of an inertial measurement device attached to a user's body part in real time and collects the measured data; A neural network analysis module that analyzes data collected through the above data collection module using a neural network analysis model to produce walking data and motion data; and An inertial measurement device-based degenerative brain disease prevention system characterized by including a gait analysis module that analyzes the above gait data and motion data.

6. In Paragraph 5, The above content provision module is, A content management unit that classifies exercise content pre-stored on the above-mentioned content provision module according to one or more criteria selected from a group consisting of the type, intensity, duration, and purpose of each content, and stores and manages it in a database. A monitoring unit that continuously collects and analyzes the above walking data and the above exercise data, evaluates the progression status and improvement status of degenerative brain disease, and regularly reports to the user, A customized exercise generation unit that analyzes the above walking data and the above exercise data, and provides customized exercise content suitable for the user from among the exercise content stored in the content management unit according to the analysis result. A health program unit that provides one or more customized protocols selected from a group consisting of diet, sleep management, meditation, psychological counseling, customized nutritional supplements, walking, and running for lifestyle improvement, based on one or more pieces of information selected from a group consisting of the degree of progression of the user's degenerative brain disease, the walking data, and the exercise data; and An inertial measurement device-based degenerative brain disease prevention system characterized by including a test content providing unit that provides test content for evaluating a user's degenerative brain disease.

7. A data collection step of measuring acceleration and angular velocity data of an inertial measuring device attached to a part of the user's body in real time and collecting the measured data; A neural network analysis step that calculates gait data by analyzing data collected through a data collection module using a neural network analysis model; A gait analysis step for analyzing the above gait data to evaluate the degree of progression of degenerative brain disease and detecting abnormal gait characteristics; and A method for providing information on the prevention of degenerative brain diseases based on an inertial measurement device, characterized by including a feedback providing step of providing real-time feedback when abnormal gait characteristics are detected.

8. In Paragraph 7, A method for providing information on the prevention of degenerative brain diseases based on an inertial measurement device, characterized in that, in the data collection step, the inertial measurement device is attached to one or more body parts selected from the group consisting of the shoulder, back, knee, and foot.

9. In Paragraph 7, The above neural network analysis step is, A method for providing information on the prevention of degenerative brain diseases based on an inertial measurement device, further comprising a reliability correction step that improves the consistency and accuracy of data collected through a data collection module during the analysis of the neural network analysis model; wherein the reliability correction step performs preprocessing including data interpolation, axis correction, outlier removal, feature scaling, and normalization processes.

10. In Paragraph 7, The above neural network analysis step is, A method for providing information on the prevention of degenerative brain diseases based on an inertial measurement device, characterized by further including the above reliability correction step; and a position correction step that subsequently provides a guide to the user to adjust the sensor attachment position.

11. Content delivery step providing customized content tailored to the user's state; A data collection step of measuring acceleration and angular velocity data of an inertial measuring device attached to a user's body part in real time and collecting the measured data; A neural network analysis step for analyzing data collected through the above data collection module using a neural network analysis model to produce walking data and motion data; and A method for providing information on the prevention of degenerative brain diseases based on an inertial measurement device, characterized by including a gait analysis step for analyzing the above gait data and exercise data.

12. In Paragraph 11, The above content provision step is, A content management step of classifying exercise content pre-stored on the above-mentioned content provision module according to one or more criteria selected from a group consisting of the type, intensity, duration, and purpose of each content, and storing and managing it in a database. A monitoring step of continuously collecting and analyzing the above walking data and the above exercise data, evaluating the progression status and improvement of the degenerative brain disease, and regularly reporting to the user, A customized exercise generation step that analyzes the user's walking data and exercise data and provides customized exercise content suitable for the user from among the exercise content stored in the content management unit according to the analyzed results. A health program stage that provides one or more customized protocols selected from a group consisting of diet, sleep management, meditation, psychological counseling, customized nutritional supplements, walking, and running for lifestyle improvement, based on one or more pieces of information selected from a group consisting of the degree of progression of the user's degenerative brain disease, the walking data, and the exercise data; and A method for providing information on the prevention of degenerative brain disease based on an inertial measurement device, characterized by including a step of providing test content for evaluating a user's degenerative brain disease.

13. A data collection step for measuring acceleration and angular velocity data of an inertial measuring device attached to a part of the user's body in real time and collecting the measured data; A neural network analysis step that calculates gait data by analyzing data collected through a data collection module using a neural network analysis model; A gait analysis step for analyzing the above gait data to evaluate the degree of progression of degenerative brain disease and detecting abnormal gait characteristics; and An early diagnosis method for degenerative brain disease based on an inertial measurement device, characterized by including a feedback providing step that provides real-time feedback when abnormal gait characteristics are detected.

14. In Paragraph 13, The above neural network analysis step is, An inertial measurement device-based early diagnosis method for degenerative brain disease, further comprising a reliability correction step that improves the consistency and accuracy of data collected through a data collection module during the analysis of the neural network analysis model; wherein the reliability correction step performs preprocessing including data interpolation, axis correction, outlier removal, feature scaling, and normalization processes.

15. In Paragraph 13, The above neural network analysis step is, An inertial measurement device-based method for early diagnosis of degenerative brain disease, characterized by further including the above reliability correction step; and a position correction step that subsequently provides a guide to the user to adjust the sensor attachment position.

16. A data collection module that collects acceleration and angular velocity data obtained through an inertial measurement unit (IMU) attached to a part of the user's body, and position information data obtained through a satellite-based positioning unit (GNSS) in real time; A neural network analysis module comprising a walking data calculation unit that calculates walking data by analyzing the acceleration and angular velocity data via a neural network, and a movement path calculation unit that calculates movement path data by analyzing the location information data via a neural network; A gait analysis module that analyzes the above gait data to generate cognitive decline data indicating the degree of progression of degenerative brain disease; A wandering determination module that determines whether the user's movement corresponds to wandering by integrating and analyzing the above movement path data and the above cognitive decline data; and A satellite-based position measurement device and an inertial measurement device-based loitering detection system comprising a feedback providing module that provides real-time feedback to a user terminal or a guardian terminal when the user's movement is determined to be loitering by the loitering determination module.

17. In Paragraph 16, The above neural network analysis module is, In the event that the satellite-based position signal is interrupted or its quality is degraded, the user's movement trajectory is estimated using an inertial-based dead reckoning method utilizing acceleration and angular velocity data collected from the inertial measurement unit (IMU), and Characterized by including a path correction unit that corrects the estimated movement trajectory using a map matching method based on map information when the above satellite-based position signal is re-received. Satellite-based positioning device and inertial measurement unit-based loitering detection system.

18. In Paragraph 16, The above neural network analysis module is, It includes a reliability correction unit and a position correction unit that improve the consistency and accuracy of data collected through the above-mentioned data collection module, The above reliability correction unit performs preprocessing including data interpolation, axis correction, outlier removal, feature scaling, and normalization processes, and The above position correction unit is characterized by analyzing data collected through the data collection module and providing a guide to the user to adjust the sensor attachment position. Satellite-based positioning device and inertial measurement unit-based loitering detection system.

19. In Paragraph 16, The above-mentioned wandering determination module includes a wandering determination unit that calculates an integrated risk level based on the deviation indicator of the movement path data and the cognitive decline data, wherein The above-mentioned wandering determination unit assigns different weights to the movement path data deviation indicator according to the level of cognitive function decline data, and is characterized by determining the user's movement as unintentional wandering when the integrated risk level exceeds a preset threshold value. Satellite-based positioning device and inertial measurement unit-based loitering detection system.

20. In Paragraph 16, The above wandering determination module is, It additionally includes a standard movement pattern generation unit that generates a standard movement pattern for each user based on the user's walking and movement path patterns by accumulating and learning the above movement path data and the above cognitive decline data. Characterized by generating a determination result regarding whether the user's current movement is intentional movement or unintentional wandering by comparing the movement path data and the cognitive impairment data with the standard movement pattern for each user. Satellite-based positioning device and inertial measurement unit-based loitering detection system.

21. A data collection step for collecting acceleration and angular velocity data obtained through an inertial measurement unit (IMU) attached to a part of the user's body, and position information data obtained through a satellite-based positioning unit (GNSS) in real time; A neural network analysis step comprising a walking data calculation process that calculates walking data by analyzing the acceleration and angular velocity data in a neural network, and a movement path calculation process that calculates movement path data by analyzing the location information data in a neural network; A gait analysis step that analyzes the above gait data to generate cognitive decline data indicating the degree of progression of a degenerative brain disease; A wandering determination step for determining whether the user's movement corresponds to wandering by integrating and analyzing the above movement path data and the above cognitive decline data; and A satellite-based position measurement device and inertial measurement device-based loitering detection method comprising: a feedback providing step that provides real-time feedback to a user terminal or a guardian terminal when the user's movement is determined to be loitering in the above loitering determination step.

22. In Paragraph 21, The above neural network analysis step is, In the event that the position signal of the position measuring device (GNSS) is interrupted or its quality deteriorates, the user's movement trajectory is estimated using an inertial-based dead reckoning method utilizing acceleration and angular velocity data collected from the inertial measurement unit (IMU), and Characterized by including the operation of a path correction unit that corrects the estimated movement trajectory using a map information-based map matching method when the above location signal is re-received. Satellite-based positioning device and inertial measurement device-based wandering detection method.

23. In Paragraph 21, The above wandering determination step is, Calculate the integrated risk level by integrating the deviation indicator of the above movement path data and the above cognitive function decline data, Different weights are assigned to the movement path data deviation indicators according to the level of the cognitive function decline data, and Characterized by including a step of determining the user's movement as unintentional wandering when the above-mentioned integrated risk level exceeds a preset threshold value. Satellite-based positioning device and inertial measurement device-based wandering detection method.

Images