A method and system for collecting daily behavior data of children with autism

Abstract

This invention relates to the technical field of behavioral analysis of children with autism, and provides a method and system for collecting daily behavioral data of children with autism. The method includes: real-time acquisition of triaxial acceleration and triaxial angular velocity from an inertial measurement unit as raw data; extraction of feature data from the raw data; judgment of the physical consistency of the raw data based on the feature data, distinguishing between real limb movement signals and sensor relative motion signals; decomposition of the raw data into a first information stream and a second information stream; identification of behavioral patterns of limb movements in children with autism based on the first information stream; extraction of features from the second information stream to obtain second information features; the second information features characterizing the intensity of sensor relative motion corresponding to the sensor relative motion signal; calculation of a behavior intensity index based on the second information features; and output of the behavior pattern and behavior intensity index. This invention improves the accuracy of collecting daily behavioral data of children with autism.

Description

Technical Field

[0001] This invention relates to the technical field of behavioral analysis of children with autism, specifically to a method and system for collecting daily behavioral data of children with autism. Background Technology

[0002] To gain a deeper understanding of the motor development characteristics of children with autism and to objectively evaluate the actual effects of various interventions, researchers and clinicians tend to use inertial measurement units (IMUs) to quantify children's movement patterns. This method acquires high-precision, more objective motion data through miniature sensors, aiming to go beyond subjective observation. IMUs are typically fixed to key areas of the child's body, such as the back of the wrist, the outside of the ankle, the sternum, or the forehead, secured with medical-grade hypoallergenic adhesive patches or elastic bandages to ensure the sensor is secure and the child is comfortable. The sensors continuously record linear acceleration and angular velocity information of the body parts and transmit it wirelessly to a central data recording system for storage and processing.

[0003] When children move freely in their natural environment, the inherent behavioral characteristics of autism spectrum disorder often lead to unpredictable and sometimes violent movements. For example, many autistic children exhibit repetitive self-stimulatory behaviors such as clapping, shaking their bodies, or spinning in place. These behaviors can exert considerable mechanical stress on the sensor's anchor points, causing adhesive patches to loosen or elastic straps to shift. Sudden, impulsive movements, even crawling, rolling, or rough play, can unintentionally cause the sensor to detach or rotate. Even a slight rotation of the sensor on the limb can drastically alter the interpretation of its acceleration and angular velocity vectors, as the sensor's coordinate system is no longer accurately aligned with the body part it represents. Existing data processing workflows, typically designed for processing clear signals from stable, fixed sensors, often lack a reliable automated mechanism to reliably detect, quantify, and correct for artifacts caused by wearing instability.

[0004] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0005] This application discloses a method and system for collecting daily behavior data of children with autism, aiming to solve the technical problems in the prior art when inertial measurement units collect daily behavior data of children with autism in natural environments. These problems include data quality degradation due to sensor wearing instability, difficulty in distinguishing between real limb movement signals and sensor relative movement signals, and the inability of existing data processing procedures to effectively detect, quantify, and correct artifacts caused by wearing instability.

[0006] The technical solution of this application is as follows:

[0007] Firstly, this application discloses a method for collecting daily behavior data of children with autism, specifically including:

[0008] The triaxial acceleration and triaxial angular velocity of the inertial measurement unit are collected in real time as raw data;

[0009] Extract kinematic and dynamic feature data from the raw data;

[0010] Based on the feature data, the physical consistency of the raw data is judged to distinguish between real limb motion signals and sensor relative motion signals.

[0011] The raw data is decomposed into a first information stream and a second information stream; the first information stream corresponds to the real limb motion signal, and the second information stream corresponds to the sensor relative motion signal.

[0012] Based on the first information flow, identify the behavioral patterns of limb movement in children with autism;

[0013] The features of the second information stream are extracted to obtain the second information features; the second information features are used to characterize the intensity of the relative motion of the sensor corresponding to the sensor relative motion signal.

[0014] Calculate the behavior intensity index based on the second information feature;

[0015] Output behavioral patterns and behavioral intensity indices.

[0016] This technical solution can effectively distinguish between real limb movement signals and sensor relative movement signals, thereby improving the accuracy and reliability of daily behavior data collection for children with autism. It can also quantify the intensity of relative movement of the sensors through a behavior intensity index, providing a more refined basis for subsequent data analysis.

[0017] Furthermore, in the above-mentioned method for collecting daily behavior data of autistic children, the step of judging the physical consistency of the raw data based on the feature data and distinguishing between real limb movement signals and sensor relative movement signals includes:

[0018] Retrieve raw data and collect data from the interface interaction perception unit;

[0019] Based on the raw data, the instantaneous amplitude change rate, frequency component distribution, and physical consistency characteristics of accelerometer and gyroscope signals were extracted.

[0020] Based on the data from the interface interaction sensing unit, instantaneous pressure, distance change rate, and waveform morphology features are extracted.

[0021] Based on the raw data and the data from the interface interaction sensing unit, determine the source of the micro-motion signal:

[0022] When the inertial measurement unit detects the vibration that reaches the target, and the accelerometer and gyroscope signals maintain physical consistency with the target, and the interface interaction sensing unit simultaneously detects that the instantaneous pressure is the target pressure or the distance change rate shows periodic fluctuations in distance, it is judged as a real limb tremor.

[0023] When the inertial measurement unit detects the vibration that has reached the target, but the physical consistency between the accelerometer and gyroscope signals is not up to standard, and at the same time the interface interaction sensing unit detects a pressure spike with instantaneous pressure exceeding the threshold or a sudden change in distance in the distance change rate, it is judged as a sensor mechanical micro-motion.

[0024] When the inertial measurement unit detects a weak vibration with energy below the preset threshold and frequency band above the preset threshold, and the physical consistency between the accelerometer and gyroscope signals is not met, and the interface interaction sensing unit detects that the instantaneous pressure is weak, non-periodic, and below the preset amplitude, or the distance change rate shows distance fluctuations, and the waveform characteristics do not conform to the characteristics of mechanical friction or impact, it is judged as skin physiological micro-movement.

[0025] When the inertial measurement unit detects vibration with an amplitude lower than the preset value and a specific frequency, and the physical consistency of the accelerometer and gyroscope signals meets the standard, but the interface interaction sensing unit only detects that the instantaneous pressure is weak and uniform or the distance change rate shows distance fluctuation, and the intensity and pattern are different from limb tremors or mechanical micro-movements, it is judged to be vibration transmitted by the external environment.

[0026] The signal components of the inertial measurement unit, which are identified as genuine limb micro-tremors, are integrated with the macroscopic limb motion signals into the first information stream;

[0027] The signal components of the inertial measurement unit, which are identified as mechanical micro-motions of the sensor, are integrated into a second information stream;

[0028] The suppression was identified as the signal components of the inertial measurement unit, which consisted of skin physiological micro-movements and externally transmitted vibrations.

[0029] This technical solution can accurately distinguish between real limb tremors, sensor mechanical micro-movements, skin physiological micro-movements, and vibrations transmitted from the external environment by fusing multi-source data and refining judgment logic. This effectively filters out noise and ensures the purity of the first and second information streams, laying a solid foundation for subsequent behavior recognition and intensity calculation.

[0030] Based on the above, this application further proposes that the steps for calculating the behavior intensity index according to the second information feature include:

[0031] When a specific behavioral pattern is identified in a child with autism, behavioral dynamic features are extracted from the corresponding real limb movement signals in the first information stream. These behavioral dynamic features include the peak angular velocity of limb movements, the instantaneous rate of change of acceleration, and the kinetic energy within a specific frequency range.

[0032] Continuously monitor the interface interaction sensing unit data. When the interface interaction sensing unit data detects periodic fluctuations with energy below a preset threshold and frequency above a preset threshold within a specific frequency range, and the inertial measurement unit data shows that the physical consistency meets the standard within the corresponding frequency range, the corresponding signal component is marked as non-behavioral related interface micro-tremor.

[0033] The behavior intensity index is calculated based on the characteristics of behavior dynamics and after excluding the contribution of non-behavior-related interface micro-tremors to the behavior intensity index. The behavior intensity index is calculated by weighting the peak angular velocity, the instantaneous rate of change of acceleration, and the kinetic energy within a specific frequency range.

[0034] This technical solution can eliminate the contribution of non-behavioral interface micro-tremors to the behavioral intensity index, making the calculation of the behavioral intensity index more accurate and more realistically reflecting the limb movement intensity of children with autism, and avoiding misjudgments caused by micro-movements at the sensor and skin interface.

[0035] In some preferred embodiments, the step of identifying the behavioral patterns of limb movements in children with autism based on the first information flow includes:

[0036] Multi-dimensional feature extraction is performed on the first information stream; the multi-dimensional features include the instantaneous amplitude of the motion signal, frequency components, smoothness of the motion trajectory, and rate of change of joint angles;

[0037] Based on instantaneous amplitude and frequency components, macroscopic actions and microscopic actions can be preliminarily distinguished.

[0038] The macroscopic and microscopic movements that were initially distinguished are further refined; the refined distinction includes analyzing the smoothness of the movement trajectory and the rate of change of joint angles.

[0039] Introducing behavioral context information, we can use this information to assist in making judgments on the results of refined differentiation.

[0040] This technical solution enables a more comprehensive and accurate identification of the motor behavior patterns of children with autism through multi-dimensional feature extraction and the introduction of behavioral context information. In particular, it improves the robustness of identification in distinguishing between macroscopic and microscopic movements and in fine-grained differentiation.

[0041] As an optional approach, the steps for identifying the behavioral patterns of limb movements in children with autism based on the first information flow include:

[0042] When a combination of kinematic features that does not match known behavioral pattern features is detected in the first information stream, the discovery process for new behavioral patterns is initiated.

[0043] Cluster analysis is performed on unmatched combinations of kinematic features to identify motion segments with similar features;

[0044] Based on the pre-defined representativeness requirements, kinematic feature sequences are extracted from motion segments with similar characteristics;

[0045] Based on the kinematic feature sequence, combined with the physiological indicators or environmental context information of autistic children when similar motor segments occur, new behavioral patterns are initially named and labeled;

[0046] Clinical staff are advised to manually verify and modify new behavioral patterns and to incorporate them into the behavioral pattern database.

[0047] In subsequent identification processes, the updated behavior pattern library is used for matching;

[0048] Based on the historical behavioral data of different children, the weights or recognition thresholds of each pattern in the behavioral pattern library are dynamically adjusted.

[0049] This technical solution enables the automatic discovery, naming, and inclusion of unknown or newly emerging behavioral patterns in children with autism. Through confirmation by clinical staff and dynamic adjustments to historical data, the behavioral pattern database can be continuously updated and optimized, improving the system's adaptability to individual differences and behavioral evolution.

[0050] Based on the above, this application further proposes that, based on the first information flow, the steps for identifying the behavioral patterns of limb movements in children with autism include:

[0051] Continuously monitor the kinematic feature distribution of each behavior pattern in the first information stream; the kinematic feature distribution includes instantaneous amplitude, frequency components, motion trajectory smoothness, and joint angle change rate;

[0052] Detect the deviation between the kinematic feature distribution of each behavior pattern and the preset feature distribution of the corresponding pattern in the current behavior pattern library;

[0053] Based on the direction and magnitude of the deviation, the recognition threshold and weight of the corresponding behavior pattern in the behavior pattern library are dynamically adjusted.

[0054] Retrieve and analyze historical behavioral data of children with autism to identify the evolutionary trends of their behavioral patterns at different developmental stages or before and after intervention;

[0055] Based on the evolution trend, the recognition thresholds and weights of each behavior pattern in the behavior pattern library are periodically updated.

[0056] This technical solution enables the behavior pattern recognition system to adapt to individual differences and changes in the behavior patterns of children with autism over time by continuously monitoring the kinematic feature distribution of behavior patterns and dynamically adjusting the recognition threshold and weights, thereby improving the accuracy and robustness of recognition.

[0057] Furthermore, in the above-mentioned method for collecting daily behavior data of autistic children, the step of detecting the deviation between the kinematic feature distribution of each behavior pattern and the preset feature distribution of the corresponding pattern in the current behavior pattern library includes:

[0058] Continuously monitor the kinematic feature distribution of the behavior pattern and calculate the instantaneous deviation between the kinematic feature distribution of the behavior pattern and the preset feature distribution;

[0059] Time series analysis is performed on the instantaneous deviation value to determine whether the duration, frequency, and amplitude of the instantaneous deviation value conform to the preset continuous deviation pattern, and the instantaneous deviation judgment result is obtained.

[0060] When the instantaneous deviation judgment result indicates that the instantaneous deviation value does not meet the preset continuous deviation mode, the instantaneous deviation value is marked as instantaneous noise fluctuation, and the instantaneous noise fluctuation is suppressed without triggering the adjustment operation of the recognition threshold and weight;

[0061] When the instantaneous deviation value continuously and regularly exceeds the preset threshold and conforms to the preset continuous deviation pattern, the instantaneous deviation value is judged as a true feature drift.

[0062] This technical solution effectively distinguishes between instantaneous noise fluctuations and true feature drift through time series analysis and the judgment of persistent deviation patterns, avoiding erroneous adjustments caused by transient noise and ensuring the stability and accuracy of dynamic adjustments to the behavior pattern library.

[0063] Based on the above, this application further proposes the following steps for retrieving and analyzing historical behavioral data of children with autism to identify the evolutionary trends of behavioral patterns in children with autism at different developmental stages or before and after intervention:

[0064] Multi-dimensional feature extraction was performed on the historical behavioral data of children with autism to obtain historical behavioral features;

[0065] Historical behavioral characteristics are aligned and synchronized over time to eliminate the impact of different data sources or differences in collection time.

[0066] Identify the correlations between historical behavioral characteristics;

[0067] Based on correlations, a dynamic evolution map is constructed; the dynamic evolution map is used to reflect the change path and intensity of historical behavioral characteristics under different influencing factors.

[0068] Based on dynamic evolution maps, we can identify the evolutionary trends of behavioral patterns in children with autism at different developmental stages or before and after intervention.

[0069] This technical solution enables the construction of dynamic evolution maps to deeply analyze the evolutionary trends of historical behavioral characteristics in children with autism, providing a more scientific and comprehensive basis for the periodic updating of the behavioral pattern database, thereby better adapting to children's growth and intervention effects.

[0070] Based on the above, this application further proposes a step of continuously monitoring the kinematic feature distribution of a behavioral pattern and calculating the instantaneous deviation value between the kinematic feature distribution of the behavioral pattern and a preset feature distribution, including:

[0071] Multi-scale analysis is performed on the kinematic feature data in the first information stream, including calculating the instantaneous deviation values ​​of instantaneous amplitude, frequency components, motion trajectory smoothness, and joint angle change rate within a preset short time window, and calculating the average deviation values ​​of instantaneous amplitude, frequency components, motion trajectory smoothness, and joint angle change rate within a preset long time window.

[0072] For specific random fluctuations that occur within a preset short time window, auxiliary judgment is made by combining data from the interface interaction sensing unit. When the inertial measurement unit detects vibrations with a frequency higher than the preset threshold and an amplitude lower than the preset threshold, and the interface interaction sensing unit simultaneously detects weak, non-periodic pressure or distance fluctuations with an amplitude lower than the preset threshold, and the corresponding waveform morphology does not conform to the characteristics of mechanical friction or impact, the corresponding fluctuation is marked as a random fluctuation caused by the child's slight movements or posture adjustments.

[0073] When calculating the instantaneous deviation value, the labeled random fluctuations are suppressed and filtered out from the calculation process of the instantaneous deviation value;

[0074] The instantaneous deviation value calculated within a preset short time window after random fluctuation suppression is fused with the corresponding average deviation value to obtain the instantaneous deviation value between the kinematic feature distribution of the behavior pattern and the preset feature distribution.

[0075] This technical solution enables more accurate calculation of instantaneous deviation values ​​through multi-scale analysis and suppression of random fluctuations. It effectively avoids interference from random fluctuations caused by minute movements or posture adjustments, thereby improving the accuracy of feature drift judgment.

[0076] Secondly, this application also discloses a data collection system for the daily behavior of children with autism, used to collect data on the daily behavior of children with autism, specifically including:

[0077] The raw data acquisition module is used to acquire the triaxial acceleration and triaxial angular velocity of the inertial measurement unit in real time as raw data;

[0078] The feature data extraction module is used to extract kinematic and dynamic feature data from the raw data.

[0079] The motion signal differentiation module is used to judge the physical consistency of the raw data based on the feature data, and to distinguish between real limb motion signals and sensor relative motion signals.

[0080] The raw data decomposition module is used to decompose the raw data into a first information stream and a second information stream; the first information stream corresponds to the real limb motion signal, and the second information stream corresponds to the sensor relative motion signal.

[0081] The behavior pattern recognition module is used to identify the behavior patterns of limb movements in children with autism based on the first information stream;

[0082] The information feature extraction module is used to extract features from the second information stream to obtain second information features; the second information features are used to characterize the intensity of the relative motion of the sensor corresponding to the sensor relative motion signal.

[0083] The intensity index calculation module is used to calculate the behavior intensity index based on the second information feature;

[0084] The pattern index output module is used to output behavioral patterns and behavioral intensity indices.

[0085] This technical solution provides a complete system that integrates data acquisition, feature extraction, signal differentiation, behavior recognition, and intensity calculation, enabling automated and high-precision acquisition and analysis of daily behavioral data of children with autism. It effectively solves the problems of data processing complexity and the limitations of manual intervention in existing technologies.

[0086] Beneficial effects

[0087] This application discloses a method for collecting daily behavioral data of children with autism. It uses real-time acquisition of triaxial acceleration and triaxial angular velocity from an inertial measurement unit (IMU) as raw data and extracts their kinematic and dynamic characteristics. Based on this, the method can judge the physical consistency of the raw data according to the feature data, thereby effectively distinguishing between real limb movement signals and sensor relative motion signals. This key step solves the technical problem in existing technologies where sensor wearing instability leads to decreased data quality and difficulty in distinguishing between real movement and artifacts. By decomposing the raw data into a first information stream corresponding to real limb movement signals and a second information stream corresponding to sensor relative motion signals, this method can identify the behavioral patterns of limb movements in children with autism based on the first information stream and extract features from the second information stream to characterize the intensity of sensor relative motion, thereby calculating a behavioral intensity index. Finally, the output behavioral patterns and behavioral intensity index provide a more objective and refined quantitative basis for clinical assessment and intervention effect analysis. Compared with existing technologies, this method can significantly improve the accuracy and reliability of data collection on the daily behavior of children with autism, effectively filter out noise introduced by sensor wearing instability, and quantify the intensity of the relative motion of the sensors. This overcomes the limitations of traditional methods in data collection in unstructured natural environments and provides strong technical support for a deeper understanding of the motor development characteristics of children with autism and the assessment of intervention measures. Attached Figure Description

[0088] Figure 1 This is a flowchart of a method for collecting daily behavior data of children with autism according to one embodiment of the present invention;

[0089] Figure 2 This is one of the flowcharts of a method for collecting daily behavior data of children with autism according to another embodiment of the present invention;

[0090] Figure 3 This is a second flowchart of a method for collecting daily behavior data of children with autism according to another embodiment of the present invention;

[0091] Figure 4 This is a system block diagram of a daily behavior data collection system for children with autism according to another embodiment of the present invention;

[0092] Explanation of reference numerals in the attached figures:

[0093] 1. Daily Behavior Data Collection System for Children with Autism; 11. Raw Data Collection Module; 12. Feature Data Extraction Module; 13. Motion Signal Differentiation Module; 14. Raw Data Decomposition Module; 15. Behavioral Pattern Recognition Module; 16. Information Feature Extraction Module; 17. Intensity Index Calculation Module; 18. Pattern Index Output Module. Detailed Implementation

[0094] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0095] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0096] Traditional methods for collecting daily behavioral data from children with autism in unstructured natural environments suffer from sensor instability, leading to artifacts such as abnormal spikes inconsistent with physiological movements or sensor detachment. This severely impacts data quality and the accuracy of subsequent analysis. Existing data processing workflows lack reliable automated mechanisms to detect, quantify, and correct these artifacts. Manual correction is time-consuming and labor-intensive, rendering many affected data segments unusable.

[0097] In response, this application proposes a method for collecting daily behavior data of children with autism, combining... Figure 1 As shown, it includes:

[0098] S1, real-time acquisition of the triaxial acceleration and triaxial angular velocity of the inertial measurement unit as raw data;

[0099] S2, extracts the kinematic and dynamic feature data from the raw data;

[0100] S3, based on the feature data, performs a judgment operation on the physical consistency of the raw data to distinguish between real limb motion signals and sensor relative motion signals;

[0101] S4, decompose the raw data into a first information stream and a second information stream; the first information stream corresponds to the real limb motion signal, and the second information stream corresponds to the sensor relative motion signal;

[0102] S5, based on the first information flow, identifies the behavioral patterns of limb movement in children with autism;

[0103] S6, extract the features of the second information stream to obtain the second information features; the second information features are used to characterize the intensity of the sensor relative motion corresponding to the sensor relative motion signal;

[0104] S7. Calculate the behavior intensity index based on the second information feature;

[0105] S8 outputs the behavior pattern and behavior intensity index.

[0106] An inertial measurement unit (IMU) is a miniature sensor, typically containing a three-axis accelerometer and a three-axis gyroscope, used to measure the linear acceleration and angular velocity of an object in three-dimensional space in real time. These sensors are usually fixed to key parts of a child's body, such as the wrist, ankle, or torso, to capture detailed information about their limb movements. Raw data refers to the unprocessed acceleration and angular velocity measurements directly output by the IMU. Kinematic data describes the geometric properties of the motion, such as displacement, velocity, and acceleration, without considering the forces causing the motion. Dynamic data involves the forces causing the motion, such as torque and energy. Physical consistency judgment refers to evaluating whether the signals conform to the laws of physics by analyzing the inherent correlation between data from different sensors (such as accelerometers and gyroscopes), thereby distinguishing between genuine limb movements and artifacts caused by the sensors themselves or wearing instability. The first information stream corresponds to the signal components confirmed as genuine limb movements after physical consistency judgment, while the second information stream corresponds to the signal components identified as relative motion of the sensors. A behavioral pattern refers to a repetitive or habitual sequence of movements with specific kinematic and dynamic characteristics exhibited by children with autism in their daily activities, such as clapping, shaking their body, or spinning in place. The behavioral intensity index is an indicator that quantifies the intensity or duration of these behavioral patterns.

[0107] In practical implementation, the method of this application first acquires the triaxial acceleration and triaxial angular velocity of the inertial measurement unit in real time as raw data. The inertial measurement unit can be worn on the limbs of a child with autism, for example, by fixing it to the back of the wrist, the outside of the ankle, the sternum, or the forehead using a medical-grade hypoallergenic adhesive patch or elastic bandage. These sensors continuously record data at a preset sampling frequency (e.g., 100Hz) and transmit the data to a data recording device via a wireless communication module (such as Bluetooth or Wi-Fi).

[0108] Subsequently, kinematic and dynamic feature data are extracted from the raw data. Kinematic features may include instantaneous velocity, instantaneous acceleration, angular velocity, angular acceleration, smoothness of the motion trajectory, and rate of change of joint angles. Dynamic features may include kinetic energy, torque, and power. For example, velocity and displacement can be estimated by integrating acceleration data, and attitude changes can be estimated by integrating angular velocity data. Fourier transform analysis of the signal's frequency components can also be used to identify periodic motion.

[0109] Next, based on the feature data, the physical consistency of the raw data is assessed to distinguish between real limb motion signals and sensor relative motion signals. For example, the correlation between accelerometer and gyroscope data can be compared under specific motion patterns. When a limb is making real movements, there is a fixed physical relationship between acceleration and angular velocity; however, this relationship may be disrupted when the sensor becomes loose or displaced. For example, if the accelerometer detects a violent vibration, but the gyroscope shows a small change in angular velocity, this may indicate that the sensor is sliding or experiencing an external impact, rather than real limb movement.

[0110] Then, the raw data is decomposed into a first information stream and a second information stream. The first information stream corresponds to the actual limb motion signal, and the second information stream corresponds to the sensor's relative motion signal. For example, signal separation techniques, such as independent component analysis or wavelet decomposition, can be used to decompose the raw data into different signal components. Based on the result of physical consistency judgment, signal components that conform to the characteristics of actual limb motion are assigned to the first information stream, and signal components that conform to the characteristics of sensor relative motion are assigned to the second information stream.

[0111] Based on the first information stream, behavioral patterns of limb movements in children with autism can be identified. For example, machine learning algorithms, such as support vector machines, neural networks, or hidden Markov models, can be used to classify the kinematic and dynamic features in the first information stream. The model can then be trained to identify preset behavioral patterns, such as clapping, shaking, spinning in place, walking, and running.

[0112] Simultaneously, features of the second information stream are extracted to obtain second information features. These second information features characterize the severity of the relative motion of the sensor corresponding to the sensor's relative motion signal. For example, the root mean square value, peak value, energy within a frequency range, or power spectral density of a specific frequency component of the second information stream can be calculated. These features can quantify the severity of sensor loosening, sliding, or impact.

[0113] Finally, a behavior intensity index is calculated based on the second information features. For example, the second information features can be compared with a preset threshold, or a comprehensive behavior intensity index can be calculated by weighted combination of multiple second information features. This index can reflect the degree to which the relative motion of the sensor affects data quality, thereby indirectly reflecting the intensity of the child's behavior or the stability of the sensor wearing.

[0114] Finally, the system outputs behavioral patterns and intensity indices. This information can be transmitted to a display device or stored in a database for further analysis and evaluation by clinicians or researchers.

[0115] Optional, combined Figure 2As shown, S3 performs a physical consistency judgment operation on the original data based on the feature data, and the steps to distinguish between real limb motion signals and sensor relative motion signals include:

[0116] S31, retrieve the original data and collect the interface interaction perception unit data;

[0117] S32, based on the original data, extracts the instantaneous amplitude change rate, frequency component distribution, and physical consistency characteristics of accelerometer and gyroscope signals;

[0118] S33, based on the interface interaction perception unit data, extracts instantaneous pressure, distance change rate and waveform morphology features;

[0119] S34, based on the raw data and the data from the interface interaction sensing unit, determine the source of the micro-motion signal:

[0120] When the inertial measurement unit detects the vibration that reaches the target, and the accelerometer and gyroscope signals maintain physical consistency with the target, and the interface interaction sensing unit simultaneously detects that the instantaneous pressure is the target pressure or the distance change rate shows periodic fluctuations in distance, it is judged as a real limb tremor.

[0121] When the inertial measurement unit detects the vibration that has reached the target, but the physical consistency between the accelerometer and gyroscope signals is not up to standard, and at the same time the interface interaction sensing unit detects a pressure spike with instantaneous pressure exceeding the threshold or a sudden change in distance in the distance change rate, it is judged as a sensor mechanical micro-motion.

[0122] When the inertial measurement unit detects a weak vibration with energy below the preset threshold and frequency band above the preset threshold, and the physical consistency between the accelerometer and gyroscope signals is not met, and the interface interaction sensing unit detects that the instantaneous pressure is weak, non-periodic, and below the preset amplitude, or the distance change rate shows distance fluctuations, and the waveform characteristics do not conform to the characteristics of mechanical friction or impact, it is judged as skin physiological micro-movement.

[0123] When the inertial measurement unit detects vibration with an amplitude lower than the preset value and a specific frequency, and the physical consistency of the accelerometer and gyroscope signals meets the standard, but the interface interaction sensing unit only detects that the instantaneous pressure is weak and uniform or the distance change rate shows distance fluctuation, and the intensity and pattern are different from limb tremors or mechanical micro-movements, it is judged to be vibration transmitted by the external environment.

[0124] S35, integrate the signal components of the inertial measurement unit that are determined to be real limb micro-tremors with the macroscopic limb motion signals into the first information stream;

[0125] S36, integrate the signal components of the inertial measurement unit that are determined to be mechanical micro-movements of the sensor into a second information stream;

[0126] S37, suppresses the signal components of the inertial measurement unit that are judged to be skin physiological micro-movements and externally transmitted vibrations.

[0127] The process involves several key aspects. First, retrieving raw data refers to obtaining triaxial acceleration and triaxial angular velocity data from the inertial measurement unit (IMU), reflecting the motion state of the monitored object. Second, acquiring interface interaction sensing unit data refers to obtaining pressure, distance, and other information from interface sensors in contact with the monitored object's limbs, providing direct evidence of the interaction between the limbs and the external environment. Based on the raw data, the IMU extracts instantaneous amplitude change rate, frequency component distribution, and physical consistency characteristics of the accelerometer and gyroscope signals, aiming to quantify the characteristics of the IMU data from a kinematic and dynamic perspective. Physical consistency characteristics are used to assess the inherent coordination between accelerometer and gyroscope data when describing the same motion and are a key indicator for judging signal reliability. Third, based on the interface interaction sensing unit data, instantaneous pressure, distance change rate, and waveform morphology characteristics are extracted to capture the specific manifestations of limb or sensor micro-movements at the interaction level. For example, instantaneous pressure reflects the intensity of contact, distance change rate reflects relative displacement, and waveform morphology characteristics reveal the periodicity or suddenness of motion.

[0128] Based on the extracted raw data and the characteristics of the interface interaction sensing unit data, the source of the micro-motion signal is determined. Specifically, when the inertial measurement unit detects vibrations that meet preset standards, and the accelerometer and gyroscope signals maintain adequate physical consistency, while the interface interaction sensing unit simultaneously detects instantaneous pressure equal to the target pressure or distance change rate exhibiting periodic fluctuations, these combined characteristics indicate that the signal originates from genuine limb micro-tremors. For example, when a child performs fine motor skills, their fingers or wrists may produce small but regular vibrations, while simultaneously generating stable contact pressure or periodic distance changes with the operating interface.

[0129] When the inertial measurement unit detects vibrations that meet preset standards, but the physical consistency between the accelerometer and gyroscope signals fails to meet the standards, and simultaneously the interface interaction sensing unit detects a pressure spike exceeding the threshold or a sudden change in distance rate, these combined characteristics typically indicate that the sensor itself has undergone minor mechanical movement. For example, the sensor may generate instantaneous, intense signals that are not perfectly synchronized with limb movements due to loosening or minor external impacts during wear.

[0130] When the inertial measurement unit detects weak vibrations with energy below a preset threshold and frequency band above a preset threshold, and the physical consistency between the accelerometer and gyroscope signals is not met, while the interface interaction sensing unit detects weak, non-periodic pressure below a preset amplitude or distance fluctuations in the instantaneous pressure or distance change rate, and the waveform characteristics do not conform to mechanical friction or impact characteristics, these situations are usually judged as skin physiological micro-movements. For example, tiny tremors on the skin surface or slight muscle contractions generate low signal strength and do not possess typical characteristics of mechanical motion.

[0131] When the inertial measurement unit (IMU) detects vibrations with amplitudes below a preset value and a specific frequency, and the physical consistency of the accelerometer and gyroscope signals meets the standard, but the interface interaction sensing unit only detects weak, uniform instantaneous pressure or distance fluctuations in the rate of change of distance, and the intensity and pattern of these fluctuations are different from limb tremors or mechanical micro-movements, these signals are judged as vibrations transmitted from the external environment. For example, low-intensity, specific-frequency background vibrations from the surrounding environment, such as the operation of an air conditioner or fan, may be captured by the IMU, but are unrelated to the movement of the limbs or sensors themselves.

[0132] Optional, combined Figure 3 As shown, S7 calculates the behavior intensity index based on the second information feature, including the following steps:

[0133] S71, when a specific behavioral pattern being performed by a child with autism is identified, behavioral dynamic features are extracted from the corresponding real limb movement signals in the first information stream; the behavioral dynamic features include the peak angular velocity of limb movement, the instantaneous rate of change of acceleration, and the motion energy within a specific frequency range;

[0134] S72 continuously monitors the interface interaction sensing unit data. When the interface interaction sensing unit data detects periodic fluctuations with energy below a preset threshold and frequency above a preset threshold within a specific frequency range, and the inertial measurement unit data shows that the physical consistency meets the standard within the corresponding frequency range, the corresponding signal component is marked as non-behavioral related interface micro-shock.

[0135] S73. Based on the behavioral dynamics characteristics and excluding the contribution of non-behavioral interface micro-tremors to the behavioral intensity index, the behavioral intensity index is calculated. The behavioral intensity index is calculated by weighting the peak angular velocity, the instantaneous rate of change of acceleration, and the kinetic energy within a specific frequency range.

[0136] When the system identifies a child with autism performing a specific behavioral pattern, such as clapping, stomping, or shaking their body, it extracts behavioral dynamics features directly related to these patterns from the first information stream. These features are indicators that directly reflect the intensity and characteristics of limb movements, specifically including peak angular velocity, instantaneous rate of change of acceleration, and kinetic energy within a specific frequency range. Peak angular velocity quantifies the intensity of rotational movements, instantaneous rate of change of acceleration reflects the explosiveness and speed of limb movements, and kinetic energy within a specific frequency range captures the energy distribution characteristics of different types of behaviors (such as repetitive stereotyped behaviors).

[0137] Meanwhile, to further improve the accuracy of the behavioral intensity index, this application continuously monitors the data from the interface interaction sensing unit. The interface interaction sensing unit can be understood as an external sensor that comes into contact with the child's body or wearable device, such as a pressure sensor or distance sensor. When this unit detects periodic fluctuations within a specific frequency range where the energy is below a preset threshold and the frequency is above a preset threshold, and the inertial measurement unit data shows adequate physical consistency within the corresponding frequency range, these signal components are labeled as non-behavioral interface tremors. These tremors are typically caused by the child's involuntary, non-behavioral interactions with the external interface (such as clothing friction or slight contact with objects), rather than actual limb movements. By identifying and labeling these tremors, they can be effectively distinguished from actual limb movements.

[0138] Therefore, when calculating the Behavioral Intensity Index, the extracted behavioral dynamics features are comprehensively considered, and the contribution of labeled, non-behavioral-related interface tremors to the index is excluded. Specifically, the Behavioral Intensity Index is calculated by weighting peak angular velocity, instantaneous rate of change of acceleration, and kinetic energy within a specific frequency range. Different weighting configurations can be used for different behavioral patterns or assessment needs to highlight the importance of specific features. In this way, the calculated Behavioral Intensity Index can more accurately reflect the true behavioral intensity of children with autism and reduce external interference.

[0139] Optionally, based on the first information stream, the steps for identifying behavioral patterns of limb movement in children with autism include:

[0140] Multi-dimensional feature extraction is performed on the first information stream; the multi-dimensional features include the instantaneous amplitude of the motion signal, frequency components, smoothness of the motion trajectory, and rate of change of joint angles;

[0141] Based on instantaneous amplitude and frequency components, macroscopic actions and microscopic actions can be preliminarily distinguished.

[0142] The macroscopic and microscopic movements that were initially distinguished are further refined; the refined distinction includes analyzing the smoothness of the movement trajectory and the rate of change of joint angles.

[0143] Introducing behavioral context information, we can use this information to assist in making judgments on the results of refined differentiation.

[0144] Specifically, multi-dimensional feature extraction of the first information stream refers to obtaining multiple independent or related data dimensions from processed real limb motion signals that can comprehensively characterize the limb motion state. Among these, the instantaneous amplitude of the motion signal can be understood as the intensity or magnitude of the limb motion at a certain moment, such as the instantaneous vector magnitude of acceleration or angular velocity detected by an inertial measurement unit; frequency components refer to the energy distribution of the motion signal within different frequency ranges, such as the spectral characteristics obtained through signal processing methods like Fourier transform, aiming to reveal the periodicity or rhythmicity of the motion; the smoothness of the motion trajectory refers to the continuity and stability of the limb motion path, such as by calculating the trajectory curvature or rate of change of velocity, aiming to distinguish between smooth movements and stiff or uncoordinated movements; the rate of change of joint angles refers to the rotational speed of the limb joints per unit time, such as by differential calculation of joint angles obtained after attitude calculation from inertial measurement unit data, aiming to reflect the precision and coordination of limb movements.

[0145] Furthermore, the preliminary distinction between macroscopic and microscopic movements based on instantaneous amplitude and frequency components refers to using the significant differences in these two characteristics to initially classify large-amplitude, low-frequency overall limb movements (such as walking, running, and jumping) into small-amplitude, high-frequency, or irregular localized limb movements (such as hand tremors or facial twitches). Macroscopic movements are typically characterized by larger instantaneous amplitude and lower frequency components, while microscopic movements are characterized by smaller instantaneous amplitude and potentially higher or more complex frequency components.

[0146] Building upon this foundation, a more refined distinction is made between the initially differentiated macroscopic and microscopic movements. This involves utilizing finer features, such as the smoothness of the movement trajectory and the rate of change of joint angles, to classify various movements more accurately. For example, for macroscopic movements, the smoothness of their movement trajectory can be analyzed to determine their coordination; for microscopic movements, the rate of change of joint angles can be used to identify their fineness and repeatability.

[0147] Simultaneously, incorporating behavioral context information to assist in the judgment of refined distinctions involves combining the analysis results of limb movement signals with non-kinematic information such as the child's environment, ongoing activity, and physiological state (e.g., heart rate, skin conductance) to improve the accuracy and robustness of behavior recognition. For example, the microtremors of the hand detected during a child's drawing activity may have completely different meanings than those detected during emotional excitement; incorporating contextual information allows for more accurate judgment.

[0148] In some preferred embodiments, a specific example is given below. Suppose an autistic child wears an inertial measurement unit, and its first information stream is used to identify behavioral patterns.

[0149] First, the system extracts multi-dimensional features from the first information stream. For example, when a child makes a clapping motion, it extracts a high instantaneous amplitude and specific frequency components; when a child makes a fine grasping motion, it extracts a lower instantaneous amplitude, different frequency components, lower motion trajectory smoothness, and specific joint angle change rates.

[0150] Next, based on the instantaneous amplitude and frequency components, the system initially determined that the hand slapping was a macroscopic action, while the fine grasping was a microscopic action.

[0151] Then, these initially differentiated movements are further refined. For hand slapping, the system will further analyze the smoothness of its movement trajectory. For example, if the slapping movement is stiff and disjointed, the smoothness will be low. For fine grasping, the system will analyze the rate of change of joint angles to determine the accuracy and coordination of the grasp.

[0152] Finally, behavioral context information is incorporated to aid in the judgment. For example, if the system detects fine grasping movements while a child is playing with building blocks and their physiological indicators (such as heart rate) are stable, it is judged as normal play behavior; however, if a child exhibits repetitive hand-clapping movements when feeling anxious, and their heart rate increases, it is judged as a self-stimulatory behavior based on contextual information. This multi-dimensional, hierarchical, and context-dependent recognition method can significantly improve the accuracy and clinical applicability of identifying complex behavioral patterns in children with autism.

[0153] Optionally, based on the first information stream, the steps for identifying behavioral patterns of limb movement in children with autism include:

[0154] When a combination of kinematic features that does not match known behavioral pattern features is detected in the first information stream, the discovery process for new behavioral patterns is initiated.

[0155] Cluster analysis is performed on unmatched combinations of kinematic features to identify motion segments with similar features;

[0156] Based on the pre-defined representativeness requirements, kinematic feature sequences are extracted from motion segments with similar characteristics;

[0157] Based on the kinematic feature sequence, combined with the physiological indicators or environmental context information of autistic children when similar motor segments occur, new behavioral patterns are initially named and labeled;

[0158] Clinical staff are advised to manually verify and modify new behavioral patterns and to incorporate them into the behavioral pattern database.

[0159] In subsequent identification processes, the updated behavior pattern library is used for matching;

[0160] Based on the historical behavioral data of different children, the weights or recognition thresholds of each pattern in the behavioral pattern library are dynamically adjusted.

[0161] Specifically, when the system analyzes the first information stream, if the detected combination of kinematic features (such as instantaneous amplitude, frequency components, motion trajectory smoothness, joint angle change rate, etc.) does not match any known behavioral pattern features in the current behavioral pattern library, it is considered a potential new behavioral pattern, thus initiating the new behavioral pattern discovery process. This process aims to systematically identify, define, and integrate these unknown behaviors.

[0162] Cluster analysis of unmatched kinematic feature combinations involves grouping these unmatched motion segments based on the similarity of their kinematic features. For example, clustering algorithms such as K-means and DBSCAN can be used to group segments with similar trajectories, velocities, and accelerations into one category, thereby identifying sets of motion segments with inherent consistency. The aim is to automatically discover potential behavioral pattern structures from large amounts of raw data.

[0163] Furthermore, based on pre-defined representativeness requirements, kinematic feature sequences are extracted from movement segments with similar characteristics. Representativeness requirements can be understood as key dimensions or points of focus in describing behavioral patterns in clinical practice or research; for example, they might focus on the repetitiveness, duration, and intensity variations of the behavior. Kinematic feature sequences refer to kinematic feature data arranged continuously over time, which can more completely describe the dynamic process of a behavior. The aim is to provide a structured and meaningful feature description for defining new behavioral patterns.

[0164] Furthermore, based on the kinematic sequence of characteristics, combined with physiological indicators or environmental context information of autistic children when similar motor segments occur, new behavioral patterns are initially named and labeled. Physiological indicators may include heart rate, skin conductance, body temperature, etc., while environmental context information may include time, location, surrounding people, and ongoing activities. This auxiliary information helps to more accurately understand the intrinsic motivation and external triggers of the behavior, thus providing a more clinically meaningful initial name and classification for the new behavioral patterns.

[0165] Optionally, based on the first information stream, the steps for identifying behavioral patterns of limb movement in children with autism include:

[0166] Continuously monitor the kinematic feature distribution of each behavior pattern in the first information stream; the kinematic feature distribution includes instantaneous amplitude, frequency components, smoothness of motion trajectory, and rate of change of joint angle;

[0167] Detect the deviation between the kinematic feature distribution of each behavior pattern and the preset feature distribution of the corresponding pattern in the current behavior pattern library;

[0168] Based on the direction and magnitude of the deviation, the recognition threshold and weight of the corresponding behavior pattern in the behavior pattern library are dynamically adjusted.

[0169] Retrieve and analyze historical behavioral data of children with autism to identify the evolutionary trends of their behavioral patterns at different developmental stages or before and after intervention;

[0170] Based on the evolution trend, the recognition thresholds and weights of each behavior pattern in the behavior pattern library are periodically updated.

[0171] Specifically, continuously monitoring the kinematic feature distribution of each behavioral pattern in the first information stream refers to the system acquiring and analyzing kinematic feature data extracted from real limb movement signals in real time. These kinematic feature distributions can be understood as describing the performance of a specific behavioral pattern in the kinematic dimensions. For example, instantaneous amplitude characterizes the intensity or magnitude of movement, frequency components reflect the rhythm or periodicity of movement, the smoothness of the movement trajectory indicates the fluidity or coordination of movement, and the rate of change of joint angles quantifies the speed and range of limb joint movement. Through continuous monitoring of these multi-dimensional features, subtle changes in behavioral patterns can be comprehensively captured.

[0172] The detection of deviations between the kinematic feature distribution of each behavior pattern and the preset feature distribution of the corresponding pattern in the current behavior pattern library involves comparing the real-time monitored behavior pattern features with the typical or standard features of that behavior pattern stored in the behavior pattern library. The preset feature distribution is a baseline feature range or statistical model set for each known behavior pattern in the behavior pattern library. Deviation detection aims to quantify the difference between real-time behavior and standard behavior, for example, by calculating Euclidean distance, correlation coefficients, or statistical significance tests.

[0173] In practical applications, dynamically adjusting the recognition threshold and weight of the corresponding behavioral pattern in the behavioral pattern library based on the direction and magnitude of the deviation means that when a significant deviation in the kinematic feature distribution of a behavioral pattern is detected, the system adjusts the recognition threshold (the standard for determining whether the behavior has occurred) and weight (the importance of the behavior in the overall evaluation) of that behavioral pattern in the behavioral pattern library in real time or near real time, based on the specific direction of the deviation (e.g., whether the amplitude increases or decreases, or the frequency increases or decreases) and the magnitude of the deviation (the degree of deviation). For example, if the amplitude of a certain stereotyped behavior continues to increase, its recognition threshold can be appropriately increased to avoid oversensitivity; if the frequency of a certain target behavior gradually decreases, its recognition threshold can be appropriately decreased to improve its recognition sensitivity.

[0174] Furthermore, retrieving and analyzing historical behavioral data of children with autism to identify the evolutionary trends of their behavioral patterns at different developmental stages or before and after intervention refers to the system periodically accessing and processing the child's accumulated behavioral data over a period of time, or after specific events (such as the start or end of an intervention). In-depth analysis of this historical data can reveal regular changes in behavioral patterns over time, such as the evolutionary trajectory of the intensity, frequency, or duration of a certain behavior at different ages or different intervention stages.

[0175] Therefore, periodically updating the recognition thresholds and weights of each behavior pattern in the behavior pattern library based on evolutionary trends means that the system makes regular or on-demand macro-adjustments to the recognition parameters in the behavior pattern library based on evolutionary trends identified from historical data. This periodic updating ensures that the behavior pattern library remains synchronized with children's long-term development and intervention effects, thereby maintaining the long-term accuracy and effectiveness of behavior recognition.

[0176] Optionally, the step of detecting the deviation between the kinematic feature distribution of each behavior pattern and the preset feature distribution of the corresponding pattern in the current behavior pattern library includes:

[0177] Continuously monitor the kinematic feature distribution of the behavior pattern and calculate the instantaneous deviation between the kinematic feature distribution of the behavior pattern and the preset feature distribution;

[0178] Time series analysis is performed on the instantaneous deviation value to determine whether the duration, frequency, and amplitude of the instantaneous deviation value conform to the preset continuous deviation pattern, and the instantaneous deviation judgment result is obtained.

[0179] When the instantaneous deviation judgment result indicates that the instantaneous deviation value does not meet the preset continuous deviation mode, the instantaneous deviation value is marked as instantaneous noise fluctuation, and the instantaneous noise fluctuation is suppressed without triggering the adjustment operation of the recognition threshold and weight;

[0180] When the instantaneous deviation value continuously and regularly exceeds the preset threshold and conforms to the preset continuous deviation pattern, the instantaneous deviation value is judged as a true feature drift.

[0181] Specifically, continuous monitoring of the kinematic feature distribution of behavioral patterns refers to the system acquiring kinematic feature data from the first information stream in real time, such as instantaneous amplitude, frequency components, smoothness of motion trajectory, and rate of change of joint angles, and comparing it with the preset feature distribution of the corresponding pattern in the behavioral pattern library to calculate the instantaneous deviation value. This instantaneous deviation value reflects the immediate difference between the current behavioral feature and the standard behavioral feature.

[0182] The time-series analysis of instantaneous deviations aims to assess their dynamic characteristics. By analyzing the duration, frequency, and magnitude of instantaneous deviations, it can be determined whether they conform to a pre-defined pattern of persistent deviation. This pre-defined pattern can be determined based on clinical experience or extensive data analysis. For example, a deviation that consistently exceeds a specific time threshold, occurs more frequently than a certain threshold within a specific time window, and has a stable magnitude within a certain range is considered potentially significant.

[0183] When the instantaneous deviation judgment result indicates that the instantaneous deviation value does not conform to the preset persistent deviation pattern, for example, if the deviation lasts only a very short time or has a very small amplitude, it is marked as an instantaneous noise fluctuation. These noise fluctuations will be suppressed, that is, they will not be used to trigger the adjustment operation of the recognition threshold and weight in the behavior pattern library, so as to avoid the system overreacting to non-substantial changes.

[0184] Conversely, when the instantaneous deviation value consistently and regularly exceeds the preset threshold, and its duration, frequency, and amplitude all conform to the preset persistent deviation pattern, it is judged as true feature drift. True feature drift represents actual changes in the behavioral patterns of children with autism, such as stable behavioral shifts caused by intervention effects, growth and development, or environmental adaptation.

[0185] In some preferred embodiments, a specific example is given below. Suppose the system is monitoring repetitive hand-slapping behavior in a child with autism. This behavior pattern has a preset kinematic feature distribution in a behavior pattern library, including specific instantaneous amplitude, frequency components, and ranges of joint angle change rates.

[0186] The system continuously monitors the initial information stream and calculates the instantaneous deviation between the current hand movement and the preset slapping behavior pattern. For example, if a child occasionally scratches their hand briefly due to itching, this may cause a brief, small deviation in the instantaneous amplitude and frequency components. The system will detect this instantaneous deviation, but through time-series analysis, it will find that the duration of the deviation is only 0.5 seconds, the frequency is single, and the amplitude does not reach the preset persistent deviation threshold. In this case, the instantaneous deviation will be marked as an instantaneous noise fluctuation and suppressed, without triggering an adjustment to the slapping behavior recognition threshold in the behavior pattern library.

[0187] However, if the frequency of a child's slapping behavior gradually increases, and this increase lasts for several minutes, the instantaneous deviation value will consistently exceed the preset threshold. Furthermore, in time-series analysis, this deviation will exhibit characteristics of long duration (e.g., exceeding 30 seconds), a steadily increasing frequency, and gradually increasing amplitude, conforming to the preset persistent deviation pattern. In this case, the system will classify this as a true feature drift and dynamically adjust the recognition threshold and weight of slapping behavior in the behavior pattern library based on the direction and magnitude of the deviation. For example, it may increase the sensitivity of recognizing the behavior or adjust the calculation method of its intensity index to more accurately reflect the actual evolution of the child's behavioral pattern. In this way, the system can effectively distinguish between accidental physical movements and clinically significant changes in behavioral patterns, thereby providing a more accurate behavioral assessment.

[0188] Optionally, retrieving and analyzing historical behavioral data of children with autism to identify the evolutionary trends of their behavioral patterns at different developmental stages or before and after intervention includes the following steps:

[0189] Multi-dimensional feature extraction was performed on the historical behavioral data of children with autism to obtain historical behavioral features.

[0190] Multidimensional feature extraction refers to extracting information from raw historical behavioral data, such as inertial measurement unit data, interface interaction perception unit data, and other possible physiological or environmental data, to identify multiple aspects that characterize children's behavioral patterns, such as kinematic features, dynamic features, interaction features, and physiological response features. The extraction of these features aims to provide a rich and comprehensive data foundation for subsequent analysis.

[0191] Historical behavioral characteristics are aligned and synchronized over time to eliminate the impact of different data sources or differences in collection time.

[0192] In practical applications, historical behavioral data may originate from different sensors, be collected at different times, or have inconsistent sampling frequencies. Time series alignment and synchronization processes aim to unify these heterogeneous data along the time dimension, ensuring accurate temporal correspondences between different features, thereby providing a consistent data view for subsequent correlation analysis.

[0193] Identify the correlations between historical behavioral characteristics.

[0194] Specifically, correlation identification can employ statistical methods, machine learning algorithms, or pattern recognition techniques to analyze whether there are statistical correlations, causal relationships, or common trends of change among different historical behavioral characteristics. For example, it can analyze the relationship between the frequency of occurrence of specific limb movement patterns and changes in environmental stimuli or physiological indicators.

[0195] Based on correlations, a dynamic evolution map is constructed. This dynamic evolution map is used to reflect the changing paths and intensity of historical behavioral characteristics under different influencing factors.

[0196] In this context, a dynamic evolution map can be understood as a visual or model-based representation that intuitively shows how behavioral patterns evolve over time and to what extent factors (such as interventions, developmental stages, and environmental changes) influence this evolution. This map can take the form of a network diagram, state transition diagram, or multidimensional time series diagram.

[0197] Based on dynamic evolution maps, we can identify the evolutionary trends of behavioral patterns in children with autism at different developmental stages or before and after intervention.

[0198] By analyzing dynamic evolution maps, we can clearly observe how the intensity, frequency, duration, and other characteristics of specific behavioral patterns change as children grow or with intervention, thereby identifying evolutionary trends such as improvement, deterioration, stabilization, or the emergence of new behaviors.

[0199] Optionally, the step of continuously monitoring the kinematic feature distribution of the behavior pattern and calculating the instantaneous deviation of the kinematic feature distribution of the behavior pattern from the preset feature distribution includes:

[0200] Multi-scale analysis is performed on the kinematic feature data in the first information stream, including calculating the instantaneous deviation values ​​of instantaneous amplitude, frequency components, motion trajectory smoothness, and joint angle change rate within a preset short time window, and calculating the average deviation values ​​of instantaneous amplitude, frequency components, motion trajectory smoothness, and joint angle change rate within a preset long time window.

[0201] For specific random fluctuations that occur within a preset short time window, auxiliary judgment is made by combining data from the interface interaction sensing unit. When the inertial measurement unit detects vibrations with a frequency higher than the preset threshold and an amplitude lower than the preset threshold, and the interface interaction sensing unit simultaneously detects weak, non-periodic pressure or distance fluctuations with an amplitude lower than the preset threshold, and the corresponding waveform morphology does not conform to the characteristics of mechanical friction or impact, the corresponding fluctuation is marked as a random fluctuation caused by the child's slight movements or posture adjustments.

[0202] When calculating the instantaneous deviation value, the labeled random fluctuations are suppressed and filtered out from the calculation process of the instantaneous deviation value;

[0203] The instantaneous deviation value calculated within a preset short time window after random fluctuation suppression is fused with the corresponding average deviation value to obtain the instantaneous deviation value between the kinematic feature distribution of the behavior pattern and the preset feature distribution.

[0204] Specifically, multi-scale analysis refers to simultaneously considering kinematic characteristic data within both short and long time windows. Within the preset short time window, instantaneous deviations of instantaneous amplitude, frequency components, motion trajectory smoothness, and joint angle change rates can be calculated to capture immediate changes in behavioral patterns. Simultaneously, within the preset long time window, the average deviations of these features are calculated to reflect the long-term trends and stability of behavioral patterns.

[0205] Specifically, random fluctuations refer to non-behavioral motion signals caused by a child's unconscious, minute movements or postural adjustments. To accurately identify these random fluctuations, this application combines inertial measurement unit (IMU) data and interface interaction sensing unit (UIS) data for auxiliary judgment. When the IMU data detects vibrations with a high frequency but low amplitude, and the UIS simultaneously detects weak, non-periodic pressure or distance fluctuations with amplitudes below a preset threshold, and the waveform characteristics of these fluctuations do not conform to mechanical friction or impact characteristics, they can be marked as random fluctuations. For example, a child's slight shifting of body in a chair or lightly touching a tabletop with their fingers may generate such random fluctuations.

[0206] In practical applications, suppressing labeled random fluctuations means excluding these signal components identified as random fluctuations from the calculation of instantaneous deviation values. The purpose is to avoid interference from these non-behavioral noises on the instantaneous deviation values, thereby improving the accuracy of the deviation calculation.

[0207] Furthermore, the instantaneous deviation value calculated within a short time window after processing to suppress random fluctuations is fused with the average deviation value calculated within the corresponding long time window. Various fusion methods can be employed, such as weighted averaging or adaptive fusion. The aim is to comprehensively consider both the immediate changes and long-term trends in behavioral patterns, ensuring that the final instantaneous deviation value is sensitive to rapid changes in behavior while also being robust to occasional noise.

[0208] This application also discloses a system for collecting daily behavior data of children with autism, used to collect daily behavior data of children with autism, combined with... Figure 4 As shown, the autism children's daily behavior data collection system 1 includes:

[0209] The raw data acquisition module 11 is used to acquire the triaxial acceleration and triaxial angular velocity of the inertial measurement unit in real time as raw data;

[0210] Feature data extraction module 12 is used to extract kinematic and dynamic feature data from the raw data;

[0211] The motion signal differentiation module 13 is used to judge the physical consistency of the original data based on the feature data, and to distinguish between real limb motion signals and sensor relative motion signals.

[0212] The raw data decomposition module 14 is used to decompose the raw data into a first information stream and a second information stream; the first information stream corresponds to the real limb motion signal, and the second information stream corresponds to the sensor relative motion signal;

[0213] The behavior pattern recognition module 15 is used to recognize the behavior patterns of limb movements in children with autism based on the first information stream;

[0214] The information feature extraction module 16 is used to extract features from the second information stream to obtain second information features; the second information features are used to characterize the intensity of the relative motion of the sensor corresponding to the sensor relative motion signal.

[0215] The intensity index calculation module 17 is used to calculate the behavior intensity index based on the second information feature;

[0216] Pattern index output module 18 is used to output behavior patterns and behavior intensity indices.

[0217] Specifically, the raw data acquisition module can be one or more inertial measurement unit (IMU) sensors, which integrate a three-axis accelerometer and a three-axis gyroscope, and connect to the main processing unit via a wireless communication interface (such as Bluetooth or Wi-Fi). This module is responsible for acquiring real-time motion data of the child's limbs. As one implementation, the inertial measurement unit can be designed as a small, lightweight, and low-power wearable device, which is fixed to key parts of the child's body using medical-grade adhesive patches or elastic straps.

[0218] The feature data extraction module can consist of one or more processors (such as microcontrollers, digital signal processors, or general-purpose CPUs) and corresponding software algorithms. This module receives raw data transmitted from the raw data acquisition module and performs calculations of kinematic and dynamic features. For example, pre-programmed algorithms can be used to perform operations such as integration, filtering, and Fourier transform on acceleration and angular velocity data to extract features such as instantaneous velocity, acceleration, angular velocity, kinetic energy, and frequency components.

[0219] The motion signal differentiation module can be a standalone software module running on the main processing unit, or integrated into the feature data extraction module. This module uses a pre-defined physical model and algorithms to analyze the extracted feature data and determine the physical consistency of the original data. For example, it can distinguish between real limb motion signals and sensor relative motion signals by comparing the correlation between accelerometer and gyroscope data under specific motion patterns, or by analyzing the spectral characteristics of the signals.

[0220] The raw data decomposition module can be a software component responsible for logically dividing the raw data stream into a first information stream and a second information stream based on the judgment result of the motion signal differentiation module. The first information stream contains signal components identified as actual limb movements, while the second information stream contains signal components identified as relative motion of the sensors. This decomposition can be achieved through methods such as data tagging, data stream routing, or signal reconstruction.

[0221] The behavior pattern recognition module can be a software module based on a machine learning model, running on the main processing unit. This module receives the first information stream and uses a pre-trained classifier (such as support vector machine, neural network, hidden Markov model, etc.) to analyze the kinematic and dynamic features to identify specific behavior patterns of children with autism, such as clapping, shaking, walking, etc.

[0222] The information feature extraction module can be a software module responsible for extracting features from the second information stream to characterize the relative motion intensity of the sensor. For example, this module can calculate the root mean square value, peak value, energy or power spectral density within a specific frequency range of the second information stream to quantify the degree of sensor loosening, sliding, or impact.

[0223] The intensity index calculation module can be a software module that calculates the behavior intensity index based on the second information features provided by the information feature extraction module. This index can be obtained by weighting and combining multiple second information features or by comparing them with a preset threshold, to reflect the degree of influence of the relative motion of the sensor on data quality, thereby indirectly reflecting the intensity of the child's behavior or the stability of the sensor wearing.

[0224] The pattern index output module can be a software module responsible for formatting the identified behavioral patterns and the calculated behavioral intensity index, and outputting them to the user or subsequent analysis system through a display interface (such as a screen, LED indicator), storage interface (such as an SD card, cloud storage), or communication interface (such as Bluetooth, Wi-Fi).

[0225] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for collecting daily behavior data of children with autism, characterized in that, include: The triaxial acceleration and triaxial angular velocity of the inertial measurement unit are collected in real time as raw data; Extract the kinematic and dynamic feature data from the raw data; Based on the feature data, the physical consistency of the original data is judged to distinguish between real limb motion signals and sensor relative motion signals. The raw data is decomposed into a first information stream and a second information stream; the first information stream corresponds to the real limb movement signal, and the second information stream corresponds to the sensor relative motion signal. Based on the first information stream, identify the behavioral patterns of limb movement in children with autism; The features of the second information stream are extracted to obtain the second information features; the second information features are used to characterize the intensity of the relative motion of the sensor corresponding to the sensor relative motion signal; Calculate the behavior intensity index based on the second information feature; Output the behavior pattern and the behavior intensity index; Collect data from the interface interaction sensing unit, obtain information from the interface sensors that are in contact with the limbs of the monitored object, and extract instantaneous pressure, distance change rate and waveform morphology features; Based on the original data, the instantaneous amplitude change rate, frequency component distribution, and physical consistency characteristics of the accelerometer and gyroscope signals are extracted. Based on the raw data and the data from the interface interaction sensing unit, determine the source of the micro-motion signal: When the inertial measurement unit detects the vibration that reaches the target, and the accelerometer and gyroscope signals maintain physical consistency with the target, and the interface interaction sensing unit simultaneously detects that the instantaneous pressure is the target pressure or the distance change rate shows periodic fluctuations in distance, it is judged as a real limb tremor. When the inertial measurement unit detects the vibration that has reached the target, but the physical consistency between the accelerometer and gyroscope signals is not up to standard, and at the same time the interface interaction sensing unit detects a pressure spike with instantaneous pressure exceeding the threshold or a sudden change in distance in the distance change rate, it is judged as a sensor mechanical micro-motion. When the inertial measurement unit detects a weak vibration with energy below the preset threshold and frequency band above the preset threshold, and the physical consistency between the accelerometer and gyroscope signals is not met, and the interface interaction sensing unit detects that the instantaneous pressure is weak, non-periodic, and below the preset amplitude, or the distance change rate shows distance fluctuations, and the waveform characteristics do not conform to the characteristics of mechanical friction or impact, it is judged as skin physiological micro-movement. When the inertial measurement unit detects vibrations with an amplitude lower than the preset value and a specific frequency, and the physical consistency between the accelerometer and gyroscope signals meets the standard, but the interface interaction sensing unit only detects that the instantaneous pressure is weak and uniform or the distance change rate shows distance fluctuations, and the intensity and pattern are different from limb tremors or mechanical micro-movements, it is judged to be vibration transmitted by the external environment.

2. The method for collecting daily behavior data of autistic children according to claim 1, characterized in that, The step of calculating the behavior intensity index based on the second information feature includes: When a specific behavioral pattern is identified in a child with autism, behavioral dynamic features are extracted from the corresponding real limb movement signals in the first information stream; the behavioral dynamic features include the peak angular velocity of limb movement, the instantaneous rate of change of acceleration, and the kinetic energy within a specific frequency range. The interface interaction sensing unit data is continuously monitored. When the interface interaction sensing unit data detects periodic fluctuations with energy below a preset threshold and frequency above a preset threshold within a specific frequency range, and the inertial measurement unit data shows that the physical consistency meets the standard within the corresponding frequency range, the corresponding signal component is marked as non-behavioral related interface micro-tremor. Based on the aforementioned behavioral dynamics characteristics, and excluding the contribution of non-behavioral interface tremors to the behavioral intensity index, the behavioral intensity index is calculated; the behavioral intensity index is calculated by weighting and combining the peak angular velocity, the instantaneous rate of change of acceleration, and the kinetic energy within the specific frequency range. The signal components of the inertial measurement unit, which are identified as genuine limb micro-tremors, are integrated with the macroscopic limb motion signals into the first information stream; The signal components of the inertial measurement unit, which are identified as mechanical micro-motions of the sensor, are integrated into a second information stream; The suppression was identified as the signal components of the inertial measurement unit, which consisted of skin physiological micro-movements and externally transmitted vibrations.

3. The method for collecting daily behavior data of autistic children according to claim 1, characterized in that, The step of identifying the behavioral patterns of limb movements in children with autism based on the first information stream includes: Multi-dimensional feature extraction is performed on the first information stream; the multi-dimensional features include the instantaneous amplitude of the motion signal, frequency components, smoothness of the motion trajectory, and rate of change of joint angles; Based on the instantaneous amplitude and frequency components, macroscopic and microscopic actions are initially distinguished. The macroscopic and microscopic movements initially distinguished are further refined; the refined distinction includes analyzing the smoothness of the movement trajectory and the rate of change of joint angles. Behavioral context information is introduced, and the results of the refined distinction are used to assist in the judgment based on the behavioral context information.

4. The method for collecting daily behavior data of autistic children according to claim 1, characterized in that, The step of identifying the behavioral patterns of limb movements in children with autism based on the first information stream includes: When a combination of kinematic features in the first information stream that does not match the features of a known behavioral pattern is detected, the discovery process for a new behavioral pattern is initiated. Cluster analysis is performed on unmatched combinations of kinematic features to identify motion segments with similar features; Based on the pre-defined representativeness requirements, kinematic feature sequences are extracted from motion segments with similar characteristics; Based on the kinematic feature sequence, combined with the physiological indicators or environmental context information of autistic children when similar motor segments occur, new behavioral patterns are initially named and labeled; Clinical staff are advised to manually verify and modify new behavioral patterns and to incorporate them into the behavioral pattern database. In subsequent identification processes, the updated behavior pattern library is used for matching; Based on the historical behavioral data of different children, the weights or recognition thresholds of each pattern in the behavioral pattern library are dynamically adjusted.

5. The method for collecting daily behavior data of autistic children according to claim 1, characterized in that, The step of identifying the behavioral patterns of limb movements in children with autism based on the first information stream includes: Continuously monitor the kinematic feature distribution of each behavior pattern in the first information stream; the kinematic feature distribution includes instantaneous amplitude, frequency components, motion trajectory smoothness, and joint angle change rate; Detect the deviation between the kinematic feature distribution of each behavior pattern and the preset feature distribution of the corresponding pattern in the current behavior pattern library; Based on the direction and magnitude of the deviation, the recognition threshold and weight of the corresponding behavior pattern in the behavior pattern library are dynamically adjusted. Retrieve and analyze historical behavioral data of children with autism to identify the evolutionary trends of their behavioral patterns at different developmental stages or before and after intervention; Based on the aforementioned evolution trend, the recognition thresholds and weights of each behavior pattern in the behavior pattern library are periodically updated.

6. The method for collecting daily behavior data of autistic children according to claim 5, characterized in that, The step of detecting the deviation between the kinematic feature distribution of each behavior pattern and the preset feature distribution of the corresponding pattern in the current behavior pattern library includes: Continuously monitor the kinematic feature distribution of the behavior pattern and calculate the instantaneous deviation value between the kinematic feature distribution of the behavior pattern and the preset feature distribution; Time series analysis is performed on the instantaneous deviation value to determine whether the duration, frequency, and amplitude of the instantaneous deviation value conform to a preset continuous deviation pattern, thereby obtaining the instantaneous deviation judgment result; When the instantaneous deviation judgment result indicates that the instantaneous deviation value does not meet the preset continuous deviation mode, the instantaneous deviation value is marked as instantaneous noise fluctuation, and the instantaneous noise fluctuation is suppressed without triggering the adjustment operation of the recognition threshold and weight; When the instantaneous deviation value continuously and regularly exceeds the preset threshold and conforms to the preset continuous deviation pattern, the instantaneous deviation value is judged as a true feature drift.

7. A method for collecting daily behavior data of autistic children according to claim 5, characterized in that, The steps of retrieving and analyzing historical behavioral data of children with autism to identify the evolutionary trends of behavioral patterns in children with autism at different developmental stages or before and after intervention include: Multi-dimensional feature extraction was performed on the historical behavioral data of children with autism to obtain historical behavioral features; The historical behavioral characteristics are time-series aligned and synchronized to eliminate the impact of different data sources or differences in collection time. Identify the correlations between the historical behavioral features; Based on the aforementioned correlation, a dynamic evolution map is constructed; the dynamic evolution map is used to reflect the change path and intensity of historical behavioral characteristics under different influencing factors. Based on the dynamic evolution map, the evolution trend of behavioral patterns of children with autism at different developmental stages or before and after intervention is identified.

8. A method for collecting daily behavior data of autistic children according to claim 6, characterized in that, The step of continuously monitoring the kinematic feature distribution of the behavior pattern and calculating the instantaneous deviation value between the kinematic feature distribution of the behavior pattern and the preset feature distribution includes: Multi-scale analysis is performed on the kinematic feature data in the first information stream, including calculating the instantaneous deviation values ​​of instantaneous amplitude, frequency components, motion trajectory smoothness, and joint angle change rate within a preset short time window, and simultaneously calculating the average deviation values ​​of instantaneous amplitude, frequency components, motion trajectory smoothness, and joint angle change rate within a preset long time window. For specific random fluctuations that occur within a preset short time window, auxiliary judgment is made by combining data from the interface interaction sensing unit. When the inertial measurement unit detects vibrations with a frequency higher than the preset threshold and an amplitude lower than the preset threshold, and the interface interaction sensing unit simultaneously detects weak, non-periodic pressure or distance fluctuations with an amplitude lower than the preset threshold, and the corresponding waveform morphology does not conform to the characteristics of mechanical friction or impact, the corresponding fluctuation is marked as a random fluctuation caused by the child's slight movements or posture adjustments. When calculating the instantaneous deviation value, the marked random fluctuations are suppressed and filtered out from the calculation process of the instantaneous deviation value; The instantaneous deviation value calculated within a preset short time window after the random fluctuation suppression process is fused with the corresponding average deviation value to obtain the instantaneous deviation value between the kinematic feature distribution of the behavior pattern and the preset feature distribution.

9. A system for collecting daily behavior data of children with autism, used to collect daily behavior data of children with autism, characterized in that, include: The raw data acquisition module is used to acquire the triaxial acceleration and triaxial angular velocity of the inertial measurement unit in real time as raw data; The feature data extraction module is used to extract the kinematic and dynamic feature data of the raw data; The motion signal differentiation module is used to judge the physical consistency of the original data based on the feature data, and to distinguish between real limb motion signals and sensor relative motion signals. The raw data decomposition module is used to decompose the raw data into a first information stream and a second information stream; the first information stream corresponds to the real limb motion signal, and the second information stream corresponds to the sensor relative motion signal. The behavior pattern recognition module is used to identify the behavior patterns of limb movements in children with autism based on the first information stream; The information feature extraction module is used to extract features from the second information stream to obtain second information features; the second information features are used to characterize the intensity of the relative motion of the sensor corresponding to the sensor relative motion signal. The intensity index calculation module is used to calculate the behavior intensity index based on the second information feature; The pattern index output module is used to output the behavior pattern and the behavior intensity index; Collect data from the interface interaction sensing unit, obtain information from the interface sensors that are in contact with the limbs of the monitored object, and extract instantaneous pressure, distance change rate and waveform morphology features; Based on the original data, the instantaneous amplitude change rate, frequency component distribution, and physical consistency characteristics of the accelerometer and gyroscope signals are extracted. Based on the raw data and the data from the interface interaction sensing unit, determine the source of the micro-motion signal: When the inertial measurement unit detects the vibration that reaches the target, and the accelerometer and gyroscope signals maintain physical consistency with the target, and the interface interaction sensing unit simultaneously detects that the instantaneous pressure is the target pressure or the distance change rate shows periodic fluctuations in distance, it is judged as a real limb tremor. When the inertial measurement unit detects the vibration that has reached the target, but the physical consistency between the accelerometer and gyroscope signals is not up to standard, and at the same time the interface interaction sensing unit detects a pressure spike with instantaneous pressure exceeding the threshold or a sudden change in distance in the distance change rate, it is judged as a sensor mechanical micro-motion. When the inertial measurement unit detects a weak vibration with energy below the preset threshold and frequency band above the preset threshold, and the physical consistency between the accelerometer and gyroscope signals is not met, and the interface interaction sensing unit detects that the instantaneous pressure is weak, non-periodic, and below the preset amplitude, or the distance change rate shows distance fluctuations, and the waveform characteristics do not conform to the characteristics of mechanical friction or impact, it is judged as skin physiological micro-movement. When the inertial measurement unit detects vibrations with an amplitude lower than the preset value and a specific frequency, and the physical consistency between the accelerometer and gyroscope signals meets the standard, but the interface interaction sensing unit only detects that the instantaneous pressure is weak and uniform or the distance change rate shows distance fluctuations, and the intensity and pattern are different from limb tremors or mechanical micro-movements, it is judged to be vibration transmitted by the external environment.

Images