Machine learning-based fall detection system and method using same

Abstract

The present invention relates to a machine learning-based fall detection system and a method using the same, comprising: a fall information request unit that receives an abnormal signal from an inertial measurement device and requests inertial data collected over a certain period to generate feature values; a fall determination unit that includes a fall prediction model that predicts a worker's fall situation from the feature values ​​and a behavior analysis model that determines whether the worker behaves according to the fall situation; a fall state management unit that controls and manages the fall state; and a fall management unit that receives worker status information from the fall detection device, determines an abnormal situation, and directs an emergency rescue request.

Description

Technology Field

[0001] The present invention relates to a machine learning-based fall detection system capable of real-time rescue measures by collecting inertial data that measures the movement and inertia of a worker performing high-altitude work at a work site and predicting and analyzing fall accidents through an artificial intelligence model, and a method using the same. Background Technology

[0002] According to the Ministry of Employment and Labor's 2021 analysis of industrial accidents, the construction industry accounts for the largest proportion of occupational accident casualties at 26.29% compared to all industries, and the construction industry also shows the highest figure for occupational accident fatalities at 50.3%.

[0003] Furthermore, according to annual trends in industrial accident indicators, the construction industry is showing an increasing trend, indicating that it is vulnerable to safety management compared to other industries.

[0004] In particular, the construction industry has a higher proportion of work at heights compared to other industries, so falls account for the largest share of industrial accidents, and the proportion of fatal accidents among workers who suffer industrial accidents due to falls is 3.0%. Considering that deaths caused by falls account for more than half of all fatal accidents in the construction industry, it is evident that accidents caused by falls at construction sites are highly likely to lead directly to serious accidents.

[0005] Furthermore, smaller workplaces show higher accident rates, indicating that construction site workers at small and medium-sized enterprises are more vulnerable to safety management issues compared to large corporations that use internally developed safety management systems and platforms.

[0006] Since falls can occur due to a momentary mistake, prompt recognition and the establishment of response measures are essential to minimize damage; to this end, there is a need for fall safety accident management devices and systems that are affordable and easy to implement.

[0007] Conventionally, this technology is based on recognizing specific images by analyzing worker image data collected through cameras. While this allows for worker detection to recognize changes in posture or identify specific behaviors or movements, there is a problem of reduced efficiency when collecting data via external sensors or cameras due to environmental characteristics, as workers typically move between various locations to perform tasks rather than remaining in a fixed position.

[0008] In addition, wearable devices equipped with inertial sensors have a high possibility of error because they cannot clearly distinguish between a voluntary jump situation and a fall accident situation requiring rescue.

[0009] Patent Document 1 relates to a fall and fall detection system and method using a wearable sensor device, wherein it determines that a person has fallen or fallen when the change values ​​of an accelerometer, a gyroscope, and an altitude sensor fall within a pre-set danger range for a certain period of time, and can also presume that a dangerous situation has occurred if the measured change values ​​do not change for a pre-set threshold time. However, it merely determines only when the change values ​​exceed a predetermined threshold for a certain period of time and does not provide a means of prediction according to various environments or conditions. Prior art literature

[0010] Korean Published Patent Application No. 2020-0137698 The problem to be solved

[0011] To solve the above-mentioned problems, the present invention is characterized by providing a machine learning-based fall detection system capable of real-time emergency measures by analyzing whether a worker has fallen and the degree of behavior after the fall using inertial data measured from an inertial measurement device attached to a worker, and a method using the same. means of solving the problem

[0012] A machine learning-based fall detection system according to the present invention for achieving the above-mentioned purpose comprises: a fall detection device for a worker including a memory for storing commands and a processor for executing stored commands, a user interface equipped with audio, alarm and display means, and a communication module for wired and wireless communication; a fall management device that receives fall and movement information of the worker from the fall detection device, determines an abnormal situation, and directs an emergency rescue request; and a fall detection system that connects these devices to a network, wherein the fall detection device comprises: a fall information request unit that generates feature values ​​using inertial data collected over a certain period and an abnormal signal received from an inertial measurement device; a fall determination unit that includes a fall prediction model that predicts whether the worker has fallen from the feature values ​​and a movement analysis model that determines whether the worker has moved according to the fall situation; and a fall state management unit that controls and manages the fall state and periodically transmits the worker's location and movement information to the fall management device.

[0013] In addition, in a machine learning-based fall detection system according to the present invention, the fall information request unit is characterized by comprising an abnormal signal detection unit that receives an abnormal signal from an inertial measurement device and detects an abnormal situation, an inertial data collection unit that requests inertial data collected and stored for a certain period when an abnormal situation is detected, a feature generation unit that generates feature values ​​from the inertial data, and a data normalization unit that normalizes the feature values ​​to 0 to 1.

[0014] In addition, in the machine learning-based fall detection system according to the present invention, the abnormal signal is characterized by determining the case where the acceleration deviates from a predetermined threshold range.

[0015] In addition, in the machine learning-based fall detection system according to the present invention, the feature values ​​are characterized as being the maximum value, minimum value, standard deviation, absolute value, RMS, kurtosis, skewness, MMGR, DMM, and MDIF for the acceleration and angular velocity.

[0016] In addition, in the machine learning-based fall detection system according to the present invention, the fall prediction model is characterized by determining whether a fall has occurred using the feature value of the interval from the previous 1 second to the subsequent 1 second based on the abnormal situation detection point when a fall is detected by the fall determination unit.

[0017] In addition, in the machine learning-based fall detection system according to the present invention, the behavior analysis model is characterized by using the feature value of a period within at least 5 seconds based on the abnormal situation detection point when a fall is determined by the fall prediction model, and determining whether there is abnormal behavior.

[0018] In addition, in the machine learning-based fall detection system according to the present invention, the fall prediction model and the behavior analysis model are characterized as being Random Forest.

[0019] In addition, a method using a machine learning-based fall detection device according to the present invention is characterized by comprising the steps of: receiving an abnormal signal from an inertial measurement device; requesting and storing inertial data when an abnormal state is determined; generating feature values ​​of a fall section from the inertial data; determining whether a fall has occurred through a fall prediction model; generating feature values ​​of an abnormal behavior section if a fall is determined; predicting whether a worker is moving and the degree of movement through a behavior analysis model; and generating an alarm to notify nearby workers of an accident or requesting accident notification or rescue to a fall management device and an integrated control server depending on the degree of movement.

[0020] In addition, a method using a machine learning-based fall management device according to the present invention comprises the steps of: receiving a rescue request signal from a fall detection device; generating a confirmation button to generate a warning sound and check the status of a worker on the user interface of the fall detection device, and operating a timer to check whether the worker operates the button; if the timer expires or the worker presses the confirmation button, requesting dispatch to the work site and collecting information on the site situation; and requesting emergency rescue to 119 and sharing the work situation according to the degree of movement of the worker. Effects of the invention

[0021] As explained above, the machine learning-based fall detection system and the method using the same according to the present invention have the advantage of being able to detect a worker's movements in real time and take rapid emergency measures regarding a fall accident.

[0022] In addition, the machine learning-based fall detection system and method using the same according to the present invention have the advantage of being able to analyze the movement of a worker after a fall accident and notify other nearby workers of the emergency situation to respond.

[0023] Furthermore, the machine learning-based fall detection system and method using the same according to the present invention have the advantage of being able to determine various unexpected forms of fall accidents and abnormal behaviors by making predictions through an artificial intelligence-based model capable of machine learning. Brief explanation of the drawing

[0024] FIG. 1 is a fall detection system according to a preferred embodiment of the present invention. This is a drawing showing the basic configuration. FIG. 2 is a fall detection device according to a preferred embodiment of the present invention. This is a drawing showing the basic configuration. FIG. 3 is a fall information request unit according to a preferred embodiment of the present invention. It is a functional block diagram. FIG. 4 is a graph illustrating abnormal conditions measured by an inertial measuring device (M) according to a preferred embodiment of the present invention and inertial data collected. FIG. 5 is a functional block diagram of a fall detection unit according to a preferred embodiment of the present invention. FIG. 6 is a functional block diagram of a fall management device according to a preferred embodiment of the present invention. FIG. 7 is a diagram showing the operation flow of a fall detection system according to a preferred embodiment of the present invention, by operating mode. FIG. 8 is a flowchart illustrating a fall detection method using a fall detection device according to a preferred embodiment of the present invention. FIG. 9 is a flowchart illustrating a fall management method using a fall management device according to a preferred embodiment of the present invention. Specific details for implementing the invention

[0025] Embodiments that enable a person skilled in the art to easily implement the present invention are described in detail below with reference to the attached drawings. However, in describing the operating principles of preferred embodiments of the present invention in detail, if it is determined that a detailed description of related known functions or configurations may unnecessarily obscure the essence of the present invention, such detailed description is omitted.

[0026] In addition, the same reference numerals are used for parts having similar functions and operations throughout the drawings. Throughout the specification, when a part is described as being connected to another part, this includes not only cases where they are directly connected, but also cases where they are indirectly connected with other elements in between. Furthermore, the inclusion of a certain component means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Hereinafter, a machine learning-based fall detection system according to the present invention and a method using the same will be described with reference to the attached drawings.

[0028] FIG. 1 is a fall detection system according to a preferred embodiment of the present invention. This is a drawing showing the basic configuration.

[0029] Referring to FIG. 1, the fall detection system may consist of a fall detection device (100) connected to an inertial measurement device (M), a fall management device (200), a wireless network (300) connecting these devices, and an integrated control server (400) that comprehensively monitors the work status and current status of the worker.

[0030] The inertial measurement unit (M) is an Inertial Measurement Unit (IMU), a type of sensor that uses velocity, direction, gravity, and acceleration to sense the movement of a moving object, and can be attached to a part of a worker's body (mainly the back of the waist) to measure movement.

[0031] In addition, the inertial measuring device (M) can measure acceleration according to changes in movement along the X, Y, and Z axes, and can measure angular velocity, i.e., roll, pitch, and yaw, which are changes in rotation of the object through the rotation angles of the X, Y, and Z axes, and can detect critical changes in acceleration and output a separate signal in the case of abnormal behavior.

[0032] The inertial measurement device (M) can preferably transmit measured inertial data to the fall detection device (100) using wired or wireless communication, but is not limited thereto as it can be integrally mounted inside the fall detection device (100).

[0033] Additionally, the inertial measuring device (M) can measure inertial data and store it for a certain period of time, and if it detects an abnormal situation, it can transmit an abnormal signal to the fall detection device (100). Here, an abnormal situation refers to when an impact exceeding a critical range occurs (for example, when the difference in continuous acceleration signals exceeds a critical value or when free falling of 1.5 m or more).

[0034] Additionally, the inertial measurement device (M) can transmit inertial data of a certain period stored in the fall information request unit (110) when requested. However, the inertial measurement device (M) preferably generates an abnormal signal, but is not limited thereto; it may simply transmit inertial data, and the abnormal situation may be handled by the fall detection device (100).

[0035] The fall management device (200) is provided by a manager who manages a worker wearing a fall detection device (100), and can perform the function of determining whether the worker has fallen or moved based on the fall detection device (100) and dispatching to the accident site or taking urgent emergency measures.

[0036] FIG. 2 is a fall detection device (100) according to a preferred embodiment of the present invention. This is a drawing showing the basic configuration.

[0037] As illustrated in FIG. 2, the fall detection device (100) may include a fall information request unit (110) that receives an abnormal signal from an inertial measurement device (M) and requests inertial data, a fall determination unit (120) that predicts and determines the degree of a worker's fall and subsequent behavior from the collected inertial data through a machine-learned artificial intelligence model, an inertial data storage unit (130) that stores periodically collected inertial data, a fall state management unit (140) that controls and manages the worker's work status and abnormal status such as a fall, a controllable user interface (150) equipped with audio, a display, and operation buttons for mutual interaction with the worker, and a communication module (160) for mutual wired or wireless communication with the fall management device (200).

[0038] In particular, the fall status management unit (140) can periodically transmit the worker's location and movement information to the fall management device (200).

[0039] Although not illustrated in FIG. 2, the fall detection device (100) includes a memory for storing commands and a processor for executing the stored commands. It may be implemented as a wearable device that can be easily attached to a worker, but it may be implemented in various terminal forms without limitation, as a conventional portable device such as a smartphone can be utilized.

[0040] FIG. 3 is a fall information request unit (110) according to a preferred embodiment of the present invention. FIG. 4 is a functional block diagram and is a graph illustrating abnormal conditions measured by an inertial measuring device (M) according to a preferred embodiment of the present invention and inertial data collected.

[0041] Referring to FIG. 3, an abnormal signal detection unit (111) that detects an abnormal situation by receiving an abnormal signal from an inertial measurement device (M), and an inertial data collection unit (112) that requests inertial data collected and stored during a certain interval (interval before and after the occurrence of the abnormal signal) when an abnormal situation is detected.

[0042] It consists of a feature generation unit (113) that generates feature values ​​to be input to a fall judgment unit (120) from inertia data, and a data normalization unit (114) that normalizes the generated feature values ​​to values ​​between 0 and 1.

[0043] First, the abnormal signal receiving unit (111) can receive an abnormal signal generated by the inertial measuring device (M) when it determines an abnormal situation in which the acceleration due to the fall exceeds a predetermined threshold range. Here, the threshold range can be set based on the vertical downward gravitational acceleration caused by the fall.

[0044] Additionally, the inertial data collection unit (112) can collect six inertial data (acceleration and angular velocity for each X, Y, and Z axis) in a time series from the inertial measurement device (M) at regular time intervals. The regular time intervals are preferably sampled in units of 10 ms, but are not limited to this and can be sampled at various intervals.

[0045] For example, as shown in FIG. 4, the time series data collected from the inertial measuring device (M) shows a sudden change (on the Y-axis) in a specific time interval, and this interval is the starting point where an abnormal situation is detected, and at this time, the abnormal signal is transmitted to the fall detection device (100).

[0046] The feature generation unit (113) generates feature values ​​using statistical characteristics to detect and analyze specific actions and positional movements of a worker from inertial data. In the present invention, 10 statistical features can be calculated for each of the 6 inertial data as feature values ​​for detecting a fall. Then, by generating 10 feature values ​​for each of the 6 inertial data per time interval, a total of 60 feature values ​​can be generated.

[0047] These feature values ​​selected include maximum value, minimum value, standard deviation, absolute sum, RMS (Root Mean Square), kurtosis, skewness, MMGR (Gradient from maximum value to minimum value), DMM (Difference between Maximum value and Minimum value), and MDIF (Maximum for the Difference between two successive values), and a detailed explanation is as follows.

[0048] First, the maximum value indicates the maximum acceleration of the movement in terms of acceleration, and the degree of rotation can be determined by indicating the largest rotational speed in terms of angular velocity.

[0049] The minimum value represents the minimum acceleration of the movement in terms of acceleration, and in terms of angular velocity, it allows identifying cases where there is almost no rotation at the smallest rotational speed.

[0050] Standard deviation is an indicator of movement volatility, which can be used to assess the stability and consistency of the movement.

[0051] The absolute sum represents the total degree of movement in the motion.

[0052] RMS (Root Mean Square) represents the magnitude and distribution of changing values; the larger the value, the greater the likelihood of unstable and rapidly changing movements, and the smaller the value, the greater the likelihood of stable movements.

[0053] Kurtosis measures how much observations are clustered around the mean, and skewness is a measure of the asymmetry of a distribution that can indicate the intensity or speed of movement, while the kurtosis and skewness of angular velocity indicate the bias toward the direction of movement or the axis of rotation.

[0054] MMGR (Gradient from maximum value to minimum value) indicates how rapidly the movement changed.

[0055] DMM (Difference between Maximum and Minimum Value) indicates the range of change, allowing one to determine the extent of the change.

[0056] MDIF (Maximum for the Difference between two successive values) can determine how abruptly a movement has changed through the degree of change in movement.

[0057] Again in FIG. 3, the data normalization unit (114) normalizes the feature values ​​using a min-max normalization technique to scale them to a value (Xnew) between 0 and 1 as shown in the formula below. This is a means to reduce the impact caused by size differences between features, as inconsistent ranges between acquired features can affect the learning and performance of the artificial intelligence model.

[0058]

[0059] The user interface (150) may be equipped with a voice and display means for checking whether there is a fall, an alarm, and whether there is movement, and may provide a means for responding to whether there is movement by touching a specific button or a confirmation button displayed on the display screen.

[0060] The communication module (160) is sufficient if it can be wirelessly connected between the fall detection device (100) and the fall management device (200), but since various forms of communication methods can be adopted depending on the operating range, no special restrictions are placed on it.

[0061] FIG. 5 is a functional block diagram of a fall determination unit (120) according to a preferred embodiment of the present invention.

[0062] Referring to FIG. 5, the fall determination unit (120) is composed of a fall prediction model (121) that predicts whether a worker will fall, a fall determination unit (122) that makes a final determination of whether a worker will fall based on the prediction result, a movement analysis model (123) that predicts the movement state of the worker after the fall determination, and a movement determination unit (124) that makes a determination using the prediction result of the movement analysis model (123) and performs the function of generating an alarm to notify surrounding workers of the accident or requesting accident notification or rescue to the fall management device (200) and integrated control server (400) depending on the degree of movement.

[0063] Additionally, the fall determination unit (120) determines whether a fall has occurred by using the prediction result from the fall prediction model (121). If the determination result indicates a fall, the motion analysis model (123) is operated sequentially to determine the movement of the worker after the fall, and feature values ​​stored in the inertia data storage unit (130) in advance can be used as input.

[0064] First, the fall prediction model (121) takes as input the feature values ​​of six inertial data measured at a sampling period of 100 Hz (10 ms) for 2 seconds at the time of anomaly detection. For example, as shown in FIG. 4, the fall analysis section consists of data 1 second before and 1 second after the time of anomaly detection (indicated as the fall analysis section). Of course, this fall analysis section is not limited as it can be varied in many ways by a person of ordinary skill.

[0065] In addition, the prediction results of the fall prediction model (121) are classified into three labels based on the criteria for falling from a height of 1.5m or more: first, falling while tilting the upper body; second, falling while slipping the feet; and last, falling vertically.

[0066] Meanwhile, the behavior analysis model (123) takes as input the feature values ​​of six inertial data measured at a sampling period of 100 Hz (10 ms) during a period of at least 5 seconds after the point of detection of an abnormal situation.

[0067] For example, as shown in Fig. 4, it is desirable to collect data for 3 seconds after the fall analysis section (indicated as the abnormal behavior analysis section), but this is not specifically limited as it can be changed to various periods at the level of a typical technician.

[0068] In addition, the prediction results of the behavior analysis model (123) are classified into three labels of abnormal behavior in addition to four normal behavior criteria. Normal behavior is walking, going down stairs, turning in place, and sitting, and abnormal behavior is classified as first, walking while staggering or wobbling, second, walking while limping, and third, no movement.

[0069] Again in FIG. 5, the fall prediction model (121) and the behavior analysis model (123) preferably each use a Random Forest, which is a supervised learning model, but are not limited thereto, and any artificial intelligence model suitable for various anomaly detection and classification is sufficient.

[0070] In this invention, learning was performed using a dataset of approximately 1,200 cases related to falls and abnormal behaviors, and the fall-related measurement data was based on accident types established based on movement characteristics through the collection and classification of fall accident cases that frequently occur at actual construction sites.

[0071] In addition, data was obtained through drop experiments by attaching to the waist of a mannequin with human-like body segments and center of gravity, while movement data was acquired by having the experimenter perform movements with a low risk of injury.

[0072] Here, since the Random Forest model is a machine learning-based artificial intelligence model that is already publicly known or being implemented, a detailed explanation of its structure and operation is omitted.

[0073] FIG. 6 is a functional block diagram of a fall management device (200) according to a preferred embodiment of the present invention.

[0074] Referring to FIG. 6, the fall management device (200) comprises a worker status collection unit (210) that periodically collects the location of the workplace where the worker is working, the work content, and the work status from a fall detection device (100); an abnormal situation judgment unit (220) that determines the degree of abnormality by collecting whether the worker has fallen and the degree of movement from a movement judgment unit (124) within the fall detection device (100); a movement verification unit (230) that attempts to interact with the worker through the user interface of the fall detection device (100) to directly check the worker's status; and a rescue request unit (240) that requests rescue from an emergency medical institution (e.g., 119, hospital, etc.) or requests help from other workers working in a nearby location for the rescue and emergency treatment of the worker.

[0075] In particular, the movement verification unit (230) can perform a procedure to verify accidental and malfunction situations, such as when a worker accidentally drops or does not wear only the inertial measuring device (M), and can perform a procedure to verify by having the worker directly press a verification button provided on the fall detection device (100) to directly check the degree of movement.

[0076] The fall management device (200) can utilize a conventional portable device such as a smartphone, so it can be implemented in various terminal forms without being limited thereto.

[0077] FIG. 7 is a diagram showing the operation flow of a fall detection system according to a preferred embodiment of the present invention, by operating mode.

[0078] Referring to FIG. 7, the fall detection device (100) operates in three operating modes: normal mode, abnormal mode, and verification mode, and the fall management device (200) can operate in a situation management mode, an accident management mode, and an emergency management mode.

[0079] First, in the normal mode of the fall detection device (100), the fall determination unit (120) analyzes inertial data periodically collected from the inertial measurement device (M), and if no abnormality is detected for a certain period, a status (normal) signal (S1) can be periodically transmitted to the fall management device (200).

[0080] If the fall detection unit (120) determines that there is a fall and abnormal behavior, it changes from normal mode to abnormal mode and transmits a status (abnormal) signal (S2) to the fall management device (200).

[0081] The verification mode is a mode for verifying whether a worker can operate the fall detection device (100). When a status (request) signal (S3) is received from the fall management device (200), a procedure to verify the worker's status can be carried out through the user interface (150) of the fall detection device (100). The verification method may involve requesting verification from the worker via sound or an alarm. If the worker is in a state where verification is possible, a status (verification) signal (S4) may be transmitted. In the case of a minor accident or malfunction, rather than an accident, a signal to cancel the rescue request may be transmitted.

[0082] The response method is configured differently depending on the user interface (150). If it is a touchable screen, a specific icon can be clicked, and if there is no display means, a hardware-type confirmation button provided outside the fall detection device (100) can be clicked. Since this can be implemented in various forms by a person skilled in the art depending on the structure of the user interface (150), no special restrictions are placed.

[0083] Referring again to FIG. 7, the situation management mode in the fall management device (200) manages the worker's work situation through a status signal transmitted from the fall detection device (100). If a status (normal) signal is not received periodically, it can be recognized as a state of inoperability or communication error of the fall detection device (100), and subsequent measures can be taken accordingly.

[0084] If a status (abnormal) signal (S2) is received, the situation management mode is changed to an accident management mode, and a status (request) signal (S3) is transmitted to the fall management device (200). A timer is operated to check for a response, and if a status (confirmation) signal (S4) is not received within a certain period of time, the system is changed to an emergency management mode.

[0085] In emergency management mode, you can request dispatch to the work site to check with surrounding workers whether a worker is unable to move due to a fall, or in the event of an emergency, immediately notify emergency agencies via 119 to request rescue and quickly proceed with follow-up measures.

[0086] FIG. 8 is a flowchart illustrating a fall detection method using a fall detection device according to a preferred embodiment of the present invention.

[0087] Referring to FIG. 8, the fall detection device (100) comprises the steps of receiving an abnormal signal from an inertial measurement device (M) (S100), requesting and storing inertial data when an abnormal state is determined (S200), generating a feature value of the fall section from the inertial data (S300), and determining whether a fall has occurred through a fall prediction model (121) (S400).

[0088] If a fall is determined in step S400, the process may proceed to a step (S600) of generating characteristic values ​​of a pre-set abnormal behavior section to analyze whether the worker is moving and the degree of movement through a behavior analysis model (123), and a step (S700) of generating an alarm to notify nearby workers of the accident or requesting accident notification and rescue to the fall management device (200) and the integrated control server (400) depending on the degree of movement.

[0089] FIG. 9 is a flowchart illustrating a fall management method using a fall management device according to a preferred embodiment of the present invention.

[0090] Referring to FIG. 9, when the fall management device (200) receives a rescue request signal from the fall detection device (100) (S710), it generates a warning sound and a confirmation button for the operator to check the status on the user interface (150) of the rescue detection device (100), and the operator can operate a timer to check whether the confirmation button is activated (S720).

[0091] In the present invention, confirmation buttons can be displayed in various ways depending on the user interface (150). If a touchable display is provided, confirmation buttons in the form of icons ('No Problem' and 'Request for Rescue') can be created and displayed for selection. If a separate hardware-type button is provided, voice can be output to select the confirmation button.

[0092] If, after step S720, the worker does not press the confirmation button during the confirmation period and the timer expires, or if the worker presses the confirmation button, the fall management device (200) can have the manager dispatch to the work site and perform the step (S730) of collecting site situation information.

[0093] In addition, depending on the degree of movement of the worker, a step (S740) of requesting emergency rescue by 119 and sharing the work situation can be performed.

[0094] The present invention has been described above with reference to specific embodiments. However, those skilled in the art will be able to make various applications and modifications within the scope of the present invention based on the above description. Explanation of the symbols

[0095] 100: Fall detection device 110: Crash Information Request Section 111: Abnormal signal detection unit 112: Inertial Data Acquisition Unit 113: Feature generation unit 114: Data Normalization Section 120: Fall Judgment Division 121: Crash prediction model 122: Fall Judgment Unit 123: Behavior Analysis Model 124: Movement Assessment Unit 130: Inertia data storage unit 140: Fall Status Management Department 150: User Interface 160: Communication module 200: Fall Control Device 210: Worker Status Collector 220: Abnormal Situation Judgment Unit 230: Movement Confirmation Unit 240: Rescue Request Section 300: Internet 400: Integrated Control Server M: Inertial measuring device

Claims

Claim 1 A fall detection device for a worker comprising a memory for storing commands and a processor for executing stored commands, a user interface equipped with audio, a confirmation button, an alarm, and a display means, and a communication module for wired and wireless communication; a fall management device that receives fall and movement information of the worker from the fall detection device, determines an abnormal situation, and directs an emergency rescue request; and a fall detection system that connects these devices via a network, wherein the fall detection device comprises: a fall information request unit that generates feature values ​​using inertial data collected over a certain period and an abnormal signal received from an inertial measurement device; and a fall determination unit that includes a fall prediction model that predicts whether the worker has fallen from the feature values ​​and a movement analysis model that determines whether the worker has moved after the fall. The system includes a fall state management unit that controls and manages the fall state and periodically transmits the worker's location and movement information to a fall management device, wherein the fall management device comprises: a worker status collection unit that periodically collects the location of the workplace where the worker is working, the work content, and the work status from a fall detection device; an abnormal situation determination unit that determines the degree of abnormality by collecting whether the worker has fallen and the degree of movement from a movement determination unit within the fall detection device; and a movement confirmation unit that attempts to interact with the worker through the confirmation button to directly check the worker's status.A fall detection system comprising a rescue request unit that requests rescue from an emergency medical institution and requests assistance from other workers working in the vicinity of the worker for the rescue and emergency treatment of the worker, wherein the behavior analysis model utilizes the feature values ​​within a segment of at least 5 seconds from the abnormal situation detection point when a fall is determined by the fall prediction model, determines whether abnormal behavior exists using a Random Forest artificial intelligence model, and wherein the feature values ​​consist of a total of 60 values, including 10 features for each of acceleration (x,y,z) and angular velocity (roll, pitch, yaw): maximum, minimum, standard deviation, absolute value, RMS, kurtosis, skewness, MMGR, DMM, and MDIF. Claim 2 A fall detection system according to claim 1, wherein the fall information request unit comprises: an abnormal signal detection unit that receives an abnormal signal from the inertial measurement device and detects an abnormal situation; an inertial data collection unit that requests inertial data collected and stored for a certain period when the abnormal situation is detected; a feature generation unit that generates a feature value from the inertial data; and a data normalization unit that normalizes the feature value to 0 to 1. Claim 3 A fall detection system according to paragraph 2, characterized in that the abnormal signal determines when acceleration deviates from a predetermined threshold range. Claim 4 delete Claim 5 delete Claim 6 delete Claim 7 delete Claim 8 delete Claim 9 A fall management method using a fall management device of claim 1 is characterized by comprising: a step of receiving a rescue request signal from a fall detection device; a step of generating a warning sound through the user interface of the fall detection device and operating a timer to check whether a worker operates the button; a step of requesting dispatch to the work site and collecting site situation information if the timer expires or the worker presses the confirmation button for the rescue request; and a step of requesting emergency rescue to 119 and sharing the work situation according to the degree of movement of the worker.

Images