A method, system and apparatus for assessing motor and cognitive abilities

Abstract

The application discloses a kind of movement and cognitive ability evaluation method, system and device, method includes: obtaining the real-time motion information of user, including real-time joint angle information, real-time gait information, real-time attention information and real-time decision speed information;Physiology-behavior coupling model based on hidden Markov model is constructed, and joint evaluation index is generated by input real-time motion information;Real-time motion information is extracted using hierarchical feature extraction model, and motion feature vector and cognitive feature vector are obtained;Characteristic vector is fused with joint evaluation index, and abnormal state information containing movement anomaly and cognitive anomaly is output;Through dynamic threshold model detection behavior anomaly and combine preset feedback rule generates real-time feedback instruction, finally according to instruction dynamic adjustment user training task parameter.The method realizes the collaborative monitoring and adaptive training control of movement and cognitive ability by multidimensional data fusion and dynamic evaluation mechanism.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a method, system and device for evaluating the ability of motion and cognition. BACKGROUND

[0002] Traditional rehabilitation training needs to rely on hospitals or professional institutions, which is limited by site, equipment and personnel. Patients have difficulty in participating in high-frequency continuous participation. During the home period, due to the lack of real-time monitoring and guidance, the training efficiency is low, and even the dysfunction is aggravated due to the wrong action. The existing technology mainly adopts single-dimensional evaluation (such as only monitoring gait or attention), which cannot simultaneously capture the correlation of motion-cognition coordination disorder (such as the correlation of limb compensation and decision delay), and relies on fixed threshold or general evaluation model, which is difficult to adapt to the individual needs of different rehabilitation stages (such as acute stage and recovery stage). In addition, the traditional system needs to wear complex sensors or connect professional instruments, which cannot realize the convenient deployment at home, and the feedback mechanism is lagging behind, which cannot correct the abnormal posture or cognitive bias in real time during training, which easily leads to the fact that the patient's self-training at home is affected by the wrong action without being corrected in time, and even causes secondary injury. At the same time, the isolation of cognitive training also hinders the reconstruction of motion and neural function. SUMMARY

[0003] In view of the above problems, the present application provides a home rehabilitation technology which can integrate multi-modal data, dynamically evaluate the motion-cognition coupling state and real-time regulate the training parameters.

[0004] To achieve the above-mentioned purpose, in a first aspect, the present application provides a method for evaluating the ability of motion and cognition, comprising:

[0005] obtaining real-time motion information of a user, the real-time motion information comprising real-time motion data and real-time cognitive data, the real-time motion data comprising real-time joint angle information and real-time gait information, and the real-time cognitive data comprising real-time attention information and real-time decision speed information;

[0006] constructing a physiological-behavior coupling model configured to generate based on a hidden Markov model, inputting the real-time motion information into the physiological-behavior coupling model to obtain a joint evaluation index, the joint evaluation index comprising a motion complexity curve and a cognitive load index time sequence;

[0007] constructing a hierarchical feature extraction model, the hierarchical feature extraction model comprising a feature extraction unit and a feature fusion unit, inputting the real-time motion information into the feature extraction unit to obtain a motion feature vector and a cognitive feature vector;

[0008] inputting the motion feature vector, the cognitive feature vector and the joint evaluation index into the feature fusion unit to obtain abnormal state information, the abnormal state information comprising motion abnormality and cognitive abnormality.

[0009] The abnormal state information is input to the dynamic threshold model to detect the behavior anomaly of the user, and a real-time feedback instruction is generated through a preset feedback rule;

[0010] According to the real-time feedback instruction, the training task parameters of the user are adjusted in real time.

[0011] In some embodiments, a physiological-behavior coupling model is constructed, and the physiological-behavior coupling model is configured to generate, based on a hidden Markov model, a physiological-behavior coupling model including:

[0012] Obtain sample motion information, the sample motion information including sample motion data and sample cognitive data;

[0013] The sample motion information is subjected to kernel density estimation method operation to obtain joint probability distribution information, and the joint probability distribution information includes the physiological-behavior pattern of the user at a specific time point;

[0014] The joint probability distribution information is mapped to a hidden space of the hidden Markov model;

[0015] And, the length of a first preset time window is defined, and the sample motion information is divided into a plurality of first preset time windows in time sequence;

[0016] The sample motion information in each first preset time window is subjected to time alignment and normalization processing to obtain a plurality of joint observation vectors;

[0017] The plurality of joint observation vectors are arranged in time sequence and a joint observation sequence is generated to constitute an observation space of the hidden Markov model;

[0018] The transition probability matrix and the observation probability matrix of the hidden Markov model are constructed according to the hidden space and the observation space, and are trained to obtain a trained physiological-behavior coupling model;

[0019] The motion complexity curve is configured to be generated by calculating the transition frequency of the hidden space and the entropy value of the joint observation sequence, and the cognitive load index time sequence is configured to be generated by calculating the duration of the hidden space and the variance of the joint observation sequence.

[0020] In some embodiments, the motion feature vector is configured to be obtained by the following steps:

[0021] The real-time motion data is segmented according to a second preset time window to obtain a plurality of first time sequence segments, each time sequence segment corresponding to the real-time motion data in a second preset time window;

[0022] extracting a first time-domain feature for each first time sequence segment, the first time-domain feature comprising a first mean value, a first variance, a first maximum value, a first minimum value, and a first root mean square value;

[0023] extracting a first time-domain feature for each first time sequence segment, the first time-domain feature comprising a first mean value, a first variance, a first maximum value, a first minimum value, and a first root mean square value;

[0024] extracting a first time-space feature for each first time sequence segment, the first time-space feature comprising a spatial distribution of a motion trajectory, a spatial variation of a motion speed, and a spatial variation of a motion acceleration;

[0025] combining the first time-domain feature, the first frequency-domain feature, and the first time-space feature to generate a motion feature vector.

[0026] In some embodiments, the cognitive feature vector is configured to be obtained by the following steps:

[0027] segmenting the real-time cognitive data into a plurality of second time sequence segments according to a third preset time window, each second time sequence segment corresponding to real-time cognitive data within a third preset time window;

[0028] extracting a second time-domain feature for each second time sequence segment, the second time-domain feature comprising a second mean value, a second variance, a second maximum value, and a second minimum value;

[0029] extracting a second time-domain feature for each second time sequence segment, the second time-domain feature comprising a second mean value, a second variance, a second maximum value, and a second minimum value;

[0030] extracting a second time-space feature for each second time sequence segment, the second time-space feature comprising a spatial variation of an attention distribution, a spatial variation of a decision speed, and a spatial variation of a cognitive load, for representing cognitive decline signs of the user in a spatial dimension;

[0031] combining the second time-domain feature, the second frequency-domain feature, and the second time-space feature to generate a cognitive feature vector.

[0032] In some embodiments, the motion feature vector, the cognitive feature vector, and the joint evaluation index are input into a feature fusion unit to obtain the abnormal state information, comprising:

[0033] obtaining rehabilitation information of the user, the rehabilitation information comprising a rehabilitation duration and a rehabilitation progress, initializing a weight distribution rule of the motion feature vector, the cognitive feature vector, and the joint evaluation index according to the rehabilitation information, the weight distribution rule being configured as an early rehabilitation rule, a middle rehabilitation rule, and a late rehabilitation rule;

[0034] Feature alignment and normalization are performed on motion feature vectors, cognitive feature vectors, and joint evaluation indicators;

[0035] The motion feature vector, cognitive feature vector, and joint evaluation index after feature alignment and normalization are weighted and fused to obtain the fused feature vector.

[0036] The fused feature vectors are used to calculate the motion deviation index based on the motion complexity curve. The motion deviation index includes the postural deviation coefficient, joint impairment coefficient, or muscle contraction impairment coefficient.

[0037] The fused feature vectors are used to calculate the cognitive bias index based on the cognitive load index time series. The cognitive bias index includes the user's sluggishness coefficient, misunderstanding coefficient, and attention distraction coefficient during the cognitive process.

[0038] Abnormal state information is generated based on the motion deviation index and the cognitive deviation index.

[0039] In some embodiments, inputting abnormal state information into a dynamic threshold model to detect abnormal user behavior and generating real-time feedback instructions through preset feedback rules includes:

[0040] The initial thresholds of the dynamic threshold model are initialized based on rehabilitation information. The initial thresholds include motor deviation thresholds and cognitive deviation thresholds.

[0041] Based on the user's real-time motion information and real-time cognitive data, dynamically adjust the motion deviation threshold and cognitive deviation threshold, including:

[0042] The motion deviation threshold is dynamically adjusted based on the changing trend of the motion complexity curve, and the cognitive deviation threshold is dynamically adjusted based on the changing trend of the cognitive load index over time.

[0043] The motion deviation index is compared with the dynamically adjusted motion deviation threshold. If the motion deviation index exceeds the motion deviation threshold, it is judged as abnormal motion behavior.

[0044] The cognitive bias index is compared with the dynamically adjusted cognitive bias threshold. If the cognitive bias index exceeds the cognitive bias threshold, it is judged as cognitive behavior abnormality.

[0045] Detection results are generated based on abnormal motor behavior and cognitive behavior.

[0046] Generate preset feedback rules based on the test results;

[0047] Preset feedback rules include motion feedback rules and cognitive feedback rules;

[0048] Motion feedback rules are generated through the following steps:

[0049] Motor adjustment instructions are generated based on the type and degree of abnormal motor behavior. These instructions include adjusting the intensity, frequency, or postural requirements of the training task.

[0050] Cognitive feedback rules are generated through the following steps:

[0051] Cognitive adjustment instructions are generated based on the type and degree of cognitive behavioral abnormalities. These instructions include adjusting the complexity, attention requirements, or decision-making speed requirements of the training task.

[0052] By combining motion adjustment commands and cognitive adjustment commands, real-time feedback commands are generated.

[0053] In a second aspect, this application provides a motor and cognitive ability assessment system applicable to the assessment method described in the first aspect, the system comprising:

[0054] The information acquisition module is used to acquire the user's real-time motion information, which includes real-time motion data and real-time cognitive data. Real-time motion data includes real-time joint angle information and real-time gait information, while real-time cognitive data includes real-time attention information and real-time decision speed information.

[0055] The logic processing module is used to construct a physiological-behavioral coupling model, which is configured to be generated based on a hidden Markov model. Real-time motion information is input into the physiological-behavioral coupling model to obtain a joint evaluation index, which includes the motion complexity curve and the time series of the cognitive load index. A hierarchical feature extraction model is also constructed, comprising a feature extraction unit and a feature fusion unit. Real-time motion information is input into the feature extraction unit to obtain motion feature vectors and cognitive feature vectors. The motion feature vectors, cognitive feature vectors, and the joint evaluation index are input into the feature fusion unit to obtain abnormal state information, which includes motion abnormalities and cognitive abnormalities.

[0056] The anomaly feedback module is used to input abnormal status information into the dynamic threshold model to detect abnormal user behavior and generate real-time feedback instructions through preset feedback rules.

[0057] The instruction sending unit is used to adjust the user's training task parameters in real time based on real-time feedback instructions.

[0058] In a third aspect, this application also provides a motor and cognitive ability assessment device, including a smart terminal, a motion acquisition device, a cognitive acquisition device, an audio playback device, and an auxiliary correction device. The smart terminal is equipped with the assessment system described in the second aspect; the motion acquisition device is electrically connected to the smart terminal and is used to acquire real-time motion data; the cognitive acquisition device is electrically connected to the smart terminal and is used to acquire real-time cognitive data; the audio playback device is electrically connected to the smart terminal and is used to provide audio guidance to the user based on real-time feedback instructions from the smart terminal; the auxiliary correction device is electrically connected to the smart terminal and is used to correct the user's movements based on real-time feedback instructions so that the user's real-time movements meet the requirements of the training task.

[0059] In some embodiments, the smart terminal is configured as at least one of a tablet, computer, or mobile phone; the motion acquisition device is configured as at least one of a camera, camcorder, sensor mat, radar detector, or infrared sensor; and the cognitive acquisition device is configured as at least one of a smart bracelet, smart helmet, smart glasses, headphones, camera, camcorder, voice interaction device, or VR device.

[0060] In some embodiments, the sensing mat includes a pressure-sensitive capacitive matrix floor and a flexible piezoresistive material yoga mat. The pressure-sensitive capacitive matrix floor is used to capture the user's gait distribution and center of gravity trajectory in real time, and the flexible piezoresistive material yoga mat is used to detect the user's pressure distribution in zones.

[0061] Unlike existing technologies, the above technical solution has the following beneficial effects:

[0062] This invention provides a method, system, and device for assessing motor and cognitive abilities. The method includes the following steps: acquiring real-time motor information of the user, including real-time joint angle information, real-time gait information, real-time attention information, and real-time decision-making speed information; constructing a physiological-behavioral coupling model based on a Hidden Markov Model, and generating a joint assessment index containing a time series of motor complexity curves and cognitive load indices by inputting real-time motor information; extracting features from the real-time motor information using a hierarchical feature extraction model to obtain motor feature vectors and cognitive feature vectors; fusing the feature vectors with the joint assessment index to output abnormal state information containing motor abnormalities and cognitive abnormalities; detecting behavioral abnormalities using a dynamic threshold model and generating real-time feedback instructions based on preset feedback rules; and finally dynamically adjusting the user's training task parameters according to the instructions. This method achieves coordinated monitoring and adaptive training regulation of motor and cognitive abilities through multi-dimensional data fusion and dynamic assessment mechanisms.

[0063] The above description of the invention is merely an overview of the technical solution of this application. In order to enable those skilled in the art to better understand the technical solution of this application and to implement it based on the description and drawings, and to make the above-mentioned objectives and other objectives, features and advantages of this application easier to understand, the following description is provided in conjunction with the specific embodiments and drawings of this application. Attached Figure Description

[0064] The accompanying drawings are only used to illustrate the principles, implementation methods, applications, features, and effects of specific embodiments of the present invention and other related contents, and should not be considered as limitations on this application.

[0065] In the accompanying drawings of the instruction manual:

[0066] Figure 1 This is a schematic diagram illustrating steps S101 to S106 of the evaluation method described in the specific implementation embodiment;

[0067] Figure 2 This is a schematic diagram illustrating steps S201 to S207 of the evaluation method described in the specific implementation embodiment;

[0068] Figure 3 This is a schematic diagram illustrating steps S301 to S305 of the evaluation method described in the specific implementation embodiment;

[0069] Figure 4 This is a schematic diagram of the structure of the evaluation system described in a specific implementation.

[0070] Figure 5 This is a schematic diagram of the evaluation device described in a specific embodiment.

[0071] The reference numerals used in the above figures are explained as follows:

[0072] 1. Evaluation device;

[0073] 11. Evaluation system;

[0074] 111. Information Acquisition Module;

[0075] 112. Logic processing module;

[0076] 113. Anomaly Feedback Module;

[0077] 114. Command sending unit.

[0078] 12. Smart terminals;

[0079] 13. Motion capture equipment;

[0080] 14. Cognitive data acquisition equipment;

[0081] 15. Audio playback devices;

[0082] 16. Auxiliary correction equipment. Detailed Implementation

[0083] To illustrate the possible application scenarios, technical principles, implementable specific solutions, and achievable objectives and effects of this application in detail, the following description, in conjunction with the listed specific embodiments and accompanying drawings, provides a detailed explanation. The embodiments described herein are merely illustrative of the technical solutions of this application and are therefore intended to limit the scope of protection of this application.

[0084] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0085] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0086] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0087] In this application, terms such as “first” and “second” are used only to distinguish one entity or operation from another, and do not necessarily require or imply any actual quantity, hierarchy or order between these entities or operations.

[0088] Without further limitations, the use of terms such as “comprising,” “including,” “having,” or other similar open-ended expressions in this application is intended to cover non-exclusive inclusion, which does not exclude the presence of additional elements in a process, method, or product that includes the stated elements, such that a process, method, or product that includes a list of elements may include not only those defined elements but also other elements not expressly listed, or elements inherent to such a process, method, or product.

[0089] As understood in the Examination Guidelines, in this application, expressions such as "greater than," "less than," and "exceeding" are understood to exclude the stated number; expressions such as "above," "below," and "within" are understood to include the stated number. Furthermore, in the description of the embodiments in this application, "multiple" means two or more (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups" and "multiple times," unless otherwise explicitly specified.

[0090] In the description of the embodiments of this application, the space-related expressions used, such as "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "vertical," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," indicate the orientation or positional relationship based on the orientation or positional relationship shown in the specific embodiments or drawings. They are only for the purpose of describing the specific embodiments of this application or for the reader's understanding, and do not indicate or imply that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0091] The processor described in the embodiments of this application can be implemented by hardware, firmware, software, or a combination thereof. It can be a circuit, one or more of an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field-programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, or a microprocessor. It also includes other physical, biological, or chemical structures that can implement the same or equivalent functions as the processors listed above, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some or all of the steps in the computer program or method involved in the various embodiments of this application, or any combination of the steps mentioned therein.

[0092] The computer program involved in the embodiments can be stored in a computer device readable storage medium, which includes, but is not limited to, disks, magnetic tapes, magnetic cards, floppy disks, flash memory, optical disks, optical cards, read-only memory (ROM), random access memory (RAM), erasable programmable ROM (EPROM), and electrically erasable programmable ROM (EEPROM), etc., and also includes other biological, physical, or chemical structures that can achieve the same or equivalent functions as the storage media listed above, such as DNA, RNA, proteins, and other units with information storage capabilities. In specific embodiments, the storage medium involved can be one of the above-mentioned media types, or a combination of the above-mentioned media types. In different embodiments, the computer program involved in the embodiments can be centrally stored in a single medium, or distributed and stored in multiple media. The memory containing the computer device readable storage medium can be non-volatile memory or random access memory. These computer device readable storage media can be built into the device, or can be connected to the device involved in the embodiments as an external device or part of an external device. In some embodiments, the memory having a computer device readable storage medium is deployed locally; in other embodiments, the memory may be deployed remotely from the processor, for example, as a network-attached memory accessed via RF circuitry or an external port and a communication network, wherein the communication network may be the Internet, one or more intranets, a local area network (LAN), a wide area network (WLAN), a storage area network (SAN), or a suitable combination thereof, as long as computer device access to the memory is enabled. Furthermore, the computer program involved in the embodiments may be stored in plaintext / ciphertext form, or it may be designed as training data, integrated and recombined through model training and implicitly stored in the parameter states of a deep neural network or other machine learning model.

[0093] Please see Figure 1 In a first aspect, this embodiment provides a method for assessing motor and cognitive abilities, including:

[0094] S101. Obtain the user's real-time motion information, which includes real-time motion data and real-time cognitive data. The real-time motion data includes real-time joint angle information and real-time gait information. The real-time cognitive data includes real-time attention information and real-time decision speed information.

[0095] S102. Construct a physiological-behavioral coupling model. The physiological-behavioral coupling model is configured to be generated based on a hidden Markov model. Real-time motion information is input into the physiological-behavioral coupling model to obtain joint evaluation indicators, including motion complexity curves and cognitive load index time series.

[0096] S103. Construct a hierarchical feature extraction model. The hierarchical feature extraction model includes a feature extraction unit and a feature fusion unit. Real-time motion information is input into the feature extraction unit to obtain motion feature vectors and cognitive feature vectors.

[0097] S104. Input the motion feature vector, cognitive feature vector and joint evaluation index into the feature fusion unit to obtain abnormal state information, which includes motion abnormalities and cognitive abnormalities.

[0098] S105. Input abnormal status information into the dynamic threshold model to detect abnormal user behavior, and generate real-time feedback instructions through preset feedback rules.

[0099] S106. Adjust the user's training task parameters in real time based on real-time feedback instructions.

[0100] In step S101, real-time joint angle information refers to the three-dimensional spatial bending angles of the shoulder, elbow, and knee joints calculated through skeletal joint point filtering (using a Kalman filter algorithm to eliminate noise from the collected joint coordinates), used to quantify motor coordination and joint range of motion. Real-time gait information refers to the extracted cadence, stride amplitude coefficient of variation, and support phase / swing ratio, obtained through heart rate-motion correlation calibration, dynamically correcting the heart rate baseline offset based on the triaxial synthesized amplitude of the accelerometer, used to assess exercise endurance and gait stability. The formula for heart rate-motion correlation calibration is as follows:

[0101] ;

[0102] in, Heart rate measurements to be corrected; This refers to the uncorrected, raw heart rate measurement. The exercise compensation coefficient characterizes the degree of interference of a unit exercise intensity on the baseline heart rate. , , The instantaneous acceleration components of the triaxial accelerometer in volume coordinates (unit: These correspond to the intensity of movement in the forward, lateral, and vertical directions of the human body, respectively.

[0103] Real-time attention information refers to the standard deviation of fixation duration and fractal dimension of saccade paths collected by an eye tracker, used to characterize the concentration of cognitive resource allocation; real-time decision speed information refers to the average delay time from the presentation of a randomly occurring visual stimulus (such as a color change) to the user's key response in a dual-task paradigm, used to quantify information processing speed.

[0104] In step S102, the specific details of generating the physiological-behavioral coupling model based on the Hidden Markov Model are described later. After inputting real-time motion information, the hidden state sequence is obtained through Viterbi decoding. The motion complexity curve is generated by a sliding window of the approximate entropy of joint trajectories between adjacent states (reflecting the decay trend of motion control stability). The cognitive load index time series is output by weighted fusion of normalized pupil diameter (0.6 weight) and decision speed Z-score (0.4 weight) (a value >1.5 indicates high load risk).

[0105] In step S103, the hierarchical feature extraction model employs a CNN-LSTM hybrid model to extract spatiotemporal features from the raw sensor data, identifying subtle motion anomalies (such as Parkinson's tremor) and signs of cognitive decline. Real-time motion information is processed by CNN to extract spatial features, and LSTM further captures the temporal patterns of the gait cycle, ultimately outputting motion feature vectors (including stride symmetry, joint angular velocity covariance, etc.) and cognitive feature vectors (including fixation entropy, decision delay coefficient of variation, etc.). The motion feature vectors characterize the user's motion state and movement patterns, while the cognitive feature vectors characterize the user's cognitive state and cognitive load. The steps for obtaining the motion and cognitive feature vectors are detailed below.

[0106] In step S104, the feature fusion unit uses a random forest classifier. Optionally, motion abnormality is defined as a joint range of motion exceeding the user's historical data. Interval or gait periodic Fourier dominant frequency shift ,in, An interval is a criterion for identifying outliers in statistics based on the principle of normal distribution. The standard deviation is calculated based on the user's own historical joint angle data, reflecting the degree of data dispersion.

[0107] Cognitive abnormality is defined as a cognitive load index exceeding twice the individual's baseline standard deviation for five consecutive seconds. Or a sudden increase in attention span of more than 30%, among which, This represents the standard deviation calculated based on the user's own cognitive load historical baseline.

[0108] In step S105, the dynamic threshold model refers to the personalized threshold range generated by calculating the mean and standard deviation of real-time indicators based on historical user data, such as the motion complexity curve threshold = baseline mean. ,in, This represents the composite standard deviation of multi-dimensional fusion metrics based on user history training.

[0109] When abnormal status information continuously triggers the threshold for more than 3 cycles, a preset feedback rule is triggered. Optionally, the preset feedback rule activates a tiered response mechanism: Level 1 feedback (voice prompts) is activated when a single abnormality is detected, and Level 2 feedback (reducing the complexity of the training scenario) is triggered when both motion and cognitive abnormalities occur concurrently.

[0110] In step S106, the training task parameters are dynamically adjusted through real-time feedback instructions, such as extending the cognitive task interval by 200ms until the abnormal indicators return to the threshold range.

[0111] The method provided in this embodiment achieves a refined joint assessment of motor and cognitive abilities by fusing real-time motion data and cognitive data to construct a physiological-behavioral coupling model and a hierarchical feature extraction model. The motion complexity curve and cognitive load index time series generated by the Hidden Markov Model, combined with spatiotemporal features extracted by the CNN-LSTM hybrid model, can accurately capture motor coordination decline and abnormal cognitive resource allocation. The dynamic threshold model generates adaptive detection criteria based on personalized historical data, and combined with a hierarchical feedback mechanism, it can adjust training task parameters in real time when joint range of motion exceeds limits or the cognitive load index continues to exceed the standard, effectively balancing training intensity and user ability. Heart rate-motion correlation calibration eliminates the interference of physical activity on heart rate, and combined with random forest multimodal decision fusion, it significantly improves the accuracy of motion tremor recognition and cognitive overload warning, providing closed-loop adaptive regulation support for rehabilitation training and dual-task ability assessment.

[0112] Please see Figure 2 In some embodiments, a physiological-behavioral coupling model is constructed, which is configured to generate based on a hidden Markov model, including:

[0113] S201. Obtain sample motion information, which includes sample motion data and sample cognitive data;

[0114] S202. Perform kernel density estimation on the sample motion information to obtain joint probability distribution information, which includes the user's physiological-behavioral pattern at a specific time point.

[0115] S203. Map the joint probability distribution information to the hidden space of the Hidden Markov Model;

[0116] In addition, the length of the first preset time window is defined, and the sample motion information is divided into multiple first preset time windows in chronological order;

[0117] S204. Perform time alignment and normalization on the sample motion information in each first preset time window to obtain multiple joint observation vectors;

[0118] S205. Arrange multiple joint observation vectors in chronological order and generate a joint observation sequence to form the observation space of the Hidden Markov Model.

[0119] S206. Construct the transition probability matrix and observation probability matrix of the Hidden Markov Model based on the hidden space and the observation space, and train it to obtain the trained physiological-behavioral coupling model.

[0120] S207, the motion complexity curve is configured to be generated by calculating the transition frequency of the hidden space and the entropy of the joint observation sequence, and the cognitive load index time series is configured to be generated by calculating the duration of the hidden space and the variance of the joint observation sequence.

[0121] In step S201, the sample motion data includes sample joint angle information and sample gait information, while the sample cognitive data includes sample attention information and sample decision-making speed information. Preferably, the sample motion information originates from behavioral data of a specific user group under a preset experimental environment, and the number of samples is not less than [number missing]. Groups, each group of samples has a time span of [number missing]. minute.

[0122] In step S202, kernel density estimation is a nonparametric method used to estimate the probability density function. The kernel function in kernel density estimation is a Gaussian kernel function with a bandwidth of [missing information]. Determined by Silverman's rule, the calculation formula is as follows:

[0123] ;

[0124] in, The first sample standard deviation represents the degree of dispersion of data within a single feature dimension (such as speed or heart rate). The sample size is used. The bandwidth is determined using the Silverman rule. When generating the multidimensional joint probability distribution, the Gaussian kernel function for each dimension is based on its own... Adjust the smoothness level.

[0125] Joint probability distribution information is represented as a multidimensional probability density function. ,in These represent the feature dimensions of sample motion information and sample cognitive information, respectively.

[0126] In step S203, the hidden space of the Hidden Markov Model typically represents the latent states of the system. The joint probability distribution information is mapped to the hidden space of the Hidden Markov Model, thus associating the joint probability distribution information with the hidden states. The hidden space is composed of... It consists of several hidden states, each hidden state It represents the user's physiological and behavioral patterns at a specific point in time.

[0127] The mapping method is based on the state partitioning of the K-means clustering algorithm, which clusters the joint probability distribution information into... There are hidden states, and the clustering objective function is:

[0128] ;

[0129] in, For the first The center point of a hidden state.

[0130] Define the length of the first preset time window as And dynamically adjust based on the periodic characteristics of user behavior, with an adjustment range of... This is helpful for processing time series data.

[0131] In step S204, since real-time motion data and real-time cognitive data may come from different sensors or acquisition devices, there may be timestamp inconsistencies. Therefore, it is necessary to perform time alignment on the data to ensure that data within the same time window corresponds to the same moment. The real-time motion data and real-time cognitive data are normalized to eliminate dimensional differences between different features, for example, by scaling the data to the range of [0,1] or [-1,1]. The time alignment method is time warping based on linear interpolation, and the normalization method is Z-score normalization, calculated using the following formula:

[0132] ;

[0133] in, The sample mean. The second sample standard deviation represents the standard deviation of the data in each feature dimension within the current first preset time window. Each feature dimension is calculated separately within each first preset time window. For example, if heart rate data fluctuates significantly within a certain time window, then the heart rate characteristics within that window will be... The time alignment and normalization will be significantly higher than other windows. Time alignment and normalization help improve the stability and accuracy of the model, ensuring that different features (such as motion acceleration and EEG signals) within the same time window are comparable after normalization, and avoiding the impact of dimensional differences on model training.

[0134] In step S205, the observation space includes a joint observation sequence of the real-time motion data and real-time cognitive data. An observation vector is a multi-dimensional data point generated within a single time window after processing the real-time motion data and real-time cognitive data. Specifically, the observation vector is a vector composed of real-time motion data (such as joint angle information and gait information) and real-time cognitive data (such as attention information and decision speed information) after time alignment and normalization. For example, assuming that within a time window, the real-time motion data includes 3 joint angles and 2 gait features, and the real-time cognitive data includes 1 attention feature and 1 decision speed feature, then the observation vector can be represented as follows: ,in, It is real-time joint angle information. It is real-time gait information. It is real-time attention information. It provides real-time decision-making speed information. Within each first preset time window, time-aligned and normalized real-time motion data and real-time cognitive data are combined into an observation vector.

[0135] An observation sequence is a sequence of observation vectors arranged chronologically across multiple time windows, reflecting changes in a user's physiological and behavioral patterns over a continuous period. A joint observation sequence is represented as... ,in, For the first The joint observation vector across multiple time windows. By using a sliding time window approach, the observation vectors within multiple time windows are arranged in chronological order to generate an observation sequence. The observation vector describes the user's state at a specific moment, while the observation sequence describes the changes in the user's state over a period of time.

[0136] In step S206, the transition probability matrix is: This is used to describe the transition relationships between hidden states, and the observation probability matrix is... This describes the probability of generating a specific observation sequence under a given hidden state. Transition probability matrix. elements Indicates from hidden state Move to hidden state The probability is calculated using the following formula:

[0137] ;

[0138] in, From state Transition to state The frequency of;

[0139] The observation probability matrix is elements Indicates that it is in a hidden state. In observation The probability is calculated using a Gaussian distribution model:

[0140] ;

[0141] in, and They are respectively The mean vector and covariance matrix.

[0142] The hidden Markov model is trained using the Baum-Welch algorithm, with the optimization objective being to maximize the likelihood function of the observed sequence. ,in For model parameters, This represents the probability distribution of the accident state; the transition probability matrix is ​​updated iteratively during training. and observation probability matrix until the likelihood function converges.

[0143] In step S207, the motion complexity curve is generated by calculating the transition frequency of the hidden space and the entropy value of the joint observation sequence. The calculation formula is as follows:

[0144] ;

[0145] The cognitive load index time series is generated by calculating the duration of the hidden space and the variance of the joint observation sequence. The calculation formula is as follows:

[0146] ;

[0147] in, For the first The variance of the observation vector within each time window.

[0148] This embodiment achieves refined joint analysis of user movement and cognitive abilities through a physiological-behavioral coupling model construction method based on Hidden Markov Models. Kernel density estimation is used to perform nonparametric probability density estimation on sample movement and cognitive information, generating joint probability distribution information to effectively characterize the dynamic correlation of multidimensional physiological-behavioral patterns. K-means clustering is then used to map the joint probability distribution to a model derived from the model. A hidden space consisting of hidden states, combined with a dynamically adjusted first preset time window, performs time alignment and normalization on sample data to generate joint observation vectors and joint observation sequences, solving the problems of asynchronous time series and dimensional differences in multi-source sensor data. The transition probability matrix and observation probability matrix of the Hidden Markov Model are optimized based on the Baum-Welch algorithm, and the observation probability is calculated using a Gaussian distribution model to maximize the likelihood function of the observation sequence, significantly improving the model's ability to express complex behavioral patterns. The motion complexity curve quantifies the stability of action control through the entropy value of the hidden state transition frequency, and the cognitive load index time series reflects the volatility of information processing through the mean variance of the observation vector. The synergy of these two methods can accurately identify the decline in motor coordination (such as abnormal joint trajectories) and the risk of cognitive overload (such as sudden changes in decision-making delay). This method, through joint probability modeling and a dynamic time window mechanism, enhances the real-time performance and individual adaptability of physiological-behavioral pattern analysis, providing a reliable quantitative basis for capability assessment in dual-task scenarios.

[0149] Please see Figure 3 In some embodiments, the motion feature vector is configured to be obtained through the following steps:

[0150] S301. The real-time motion data is divided according to the second preset time window to obtain multiple first time series segments, and each time series segment corresponds to real-time motion data within a second preset time window.

[0151] S302. Extract the first time-domain feature for each first time series segment. The first time-domain feature includes the first mean, the first variance, the first maximum value, the first minimum value, and the first root mean square value.

[0152] S303. For each first time series segment, the time domain signal is converted into a frequency domain signal by fast Fourier transform, and the first frequency domain features are extracted. The first frequency domain features include the first dominant frequency, the first spectral energy, and the first spectral entropy.

[0153] S304. For each first time series segment, extract the first spatiotemporal features, which include the spatial distribution of the motion trajectory, the spatial variation of the motion velocity, and the spatial variation of the motion acceleration.

[0154] S305. Combine the first time-domain feature, the first frequency-domain feature, and the first spatiotemporal feature to generate a motion feature vector.

[0155] In step S302, a first time-domain feature is extracted for each first time-series segment. For example, for real-time joint angle information, a first mean value of each real-time joint angle information is calculated. and first variance :

[0156] ;

[0157] ;

[0158] in, Indicates the second preset time window. Real-time joint angle information at each moment This indicates the length of the second preset time window.

[0159] In step S303, a first frequency domain feature is extracted. The first frequency domain feature includes a first dominant frequency, a first spectral energy, and a first spectral entropy. For example, the first spectral energy of the real-time joint angle information is calculated. :

[0160] ;

[0161] in, It is the first frequency domain signal One frequency component, It is the total number of frequency components.

[0162] In step S304, the first spatiotemporal feature is used to directly characterize the user's abnormal motion signs in the spatial dimension, for example, calculating the spatial distribution of the motion trajectory of the joint angle signal. :

[0163] ;

[0164] in, It's a time delay.

[0165] In step S305, the motion feature vector can be represented as:

[0166] .

[0167] This embodiment achieves a refined representation of the user's motion state through a motion feature vector construction method that integrates multi-dimensional features. Real-time motion data is segmented based on a second preset time window, and the stability and amplitude fluctuation range of joint movements are quantified by combining the first time-domain features. Frequency-domain features are extracted using Fast Fourier Transform, which can identify frequency shifts and energy distribution anomalies in periodic motion patterns. Spatiotemporal features are represented using an autocorrelation function. The method calculates the temporal correlation of joint angles to capture spatial coordination decay (such as phase desynchronization of knee flexion-extension trajectories). It combines first-time-domain features, first-frequency-domain features, and first-spatiotemporal features to generate a multi-dimensional motion feature vector, covering the spatiotemporal characteristics of movement patterns and significantly enhancing the detection sensitivity of subtle abnormal signs (such as early tremors and disruption of gait periodicity). This method overcomes the limitations of single-dimensional analysis through multi-source feature complementarity. For example, temporal variance can verify the physical meaning of frequency domain dominant frequency shifts, and spatiotemporal distribution anomalies can explain the causes of abrupt changes in the temporal mean, providing high-resolution quantitative evidence for the early identification of motor dysfunction.

[0168] In some embodiments, the cognitive feature vector is configured to be obtained through the following steps:

[0169] The real-time cognitive data is divided into a third preset time window to obtain multiple second time series segments, and each second time series segment corresponds to a real-time cognitive data within a third preset time window.

[0170] For each second time series segment, extract the second time-domain features, which include the second mean, second variance, second maximum value, and second minimum value.

[0171] For each second time series segment, the time-domain signal is converted into a frequency-domain signal by a fast Fourier transform, and the second frequency-domain features are extracted. The second frequency-domain features include the second dominant frequency, the second spectral energy, and the second spectral entropy.

[0172] For each second time series segment, second spatiotemporal features are extracted, including spatial changes in attention distribution, spatial changes in decision speed, and spatial changes in cognitive load, which are used to characterize signs of cognitive decline in the spatial dimension.

[0173] The second time-domain feature, the second frequency-domain feature, and the second spatiotemporal feature are combined to generate a cognitive feature vector.

[0174] In this embodiment, for each second time series segment, a second time-domain feature is extracted. For example, for real-time attention information, a second mean of the real-time attention information is calculated. Second variance .

[0175] The second spatiotemporal feature, the temporal series feature, is used to directly characterize signs of cognitive decline in users in the spatial dimension.

[0176] Cognitive feature vectors can be represented as:

[0177] .

[0178] This embodiment achieves refined quantitative analysis of user cognitive states through a cognitive feature vector construction method based on multi-dimensional feature fusion. Real-time cognitive data is segmented based on a third preset time window, and the stability of cognitive ability is quantified by combining second time-domain features. Second frequency-domain features are extracted through Fast Fourier Transform, which can identify abnormal periodic patterns in cognitive activities. For example, a decrease in the spectral energy of decision-making speed may reflect disordered thinking rhythm, and an increase in the spectral entropy of attention may indicate fragmented processing of prompting information. The second spatiotemporal features are analyzed through an autocorrelation function. This method analyzes the spatial correlation of cognitive parameters to capture the spatial cumulative effects of imbalanced attentional resource allocation (such as abnormal spatial jumps in fixation points during visual tasks) or decision delays. It combines second-domain, second-frequency-domain, and second-spatiotemporal features to generate a multi-dimensional cognitive feature vector, covering the spatiotemporal characteristics of cognitive function and significantly enhancing the detection sensitivity of early cognitive decline (such as attentional drift and discretization of decision speed). Through complementary analysis of temporal fluctuation features and frequency-domain energy features, this method distinguishes between occasional distraction (sudden increase in instantaneous variance) and persistent cognitive decline (dominant frequency shift accompanied by a continuous increase in spectral entropy). Combined with spatiotemporal features, it reveals abnormal spatial allocation of cognitive resources (such as attentional distraction patterns in multi-task scenarios), providing high-resolution quantitative evidence for the early identification of cognitive impairment.

[0179] The motion feature vector and cognitive feature vector are used as the output of the feature extraction unit and passed to the feature fusion unit for subsequent abnormal state information generation.

[0180] In some embodiments, inputting motion feature vectors, cognitive feature vectors, and joint evaluation metrics into a feature fusion unit to obtain abnormal state information includes:

[0181] The system acquires the user's rehabilitation information, including rehabilitation duration and progress. Based on this information, it initializes the weighting rules for motor feature vectors, cognitive feature vectors, and joint assessment indicators. These weighting rules are configured as early rehabilitation rules, mid-term rehabilitation rules, and late-term rehabilitation rules.

[0182] Feature alignment and normalization are performed on motion feature vectors, cognitive feature vectors, and joint evaluation indicators;

[0183] The motion feature vector, cognitive feature vector, and joint evaluation index after feature alignment and normalization are weighted and fused to obtain the fused feature vector.

[0184] The fused feature vectors are used to calculate the motion deviation index based on the motion complexity curve. The motion deviation index includes the postural deviation coefficient, joint impairment coefficient, or muscle contraction impairment coefficient.

[0185] The fused feature vectors are used to calculate the cognitive bias index based on the cognitive load index time series. The cognitive bias index includes the user's sluggishness coefficient, misunderstanding coefficient, and attention distraction coefficient during the cognitive process.

[0186] Abnormal state information is generated based on the motion deviation index and the cognitive deviation index.

[0187] In this embodiment, abnormal state information is generated through dynamic weight allocation and multimodal feature fusion. Weight allocation rules are initialized based on the user's rehabilitation information. Early rehabilitation rules assign higher weights to motor feature vectors to enhance basic movement monitoring; mid-stage rehabilitation rules balance the weights of motor and cognitive features to reflect the synergy of the two tasks; and late-stage rehabilitation rules emphasize cognitive feature vectors and joint assessment indicators to evaluate higher-order functional recovery.

[0188] Feature alignment eliminates temporal biases between motion and cognitive data through dynamic time warping (DTW), and normalization uses Z-score standardization to eliminate dimensional differences in cross-modal features such as acceleration variance, joint range of motion, and response error rate.

[0189] The fusion feature vector generated after weighted fusion dynamically calculates the posture deviation coefficient (such as the Hausdorff distance between the shoulder joint trajectory and the standard rehabilitation path), joint impairment coefficient (such as the deviation of the knee joint range of motion from the preset range) or muscle contraction impairment coefficient based on the motion complexity curve.

[0190] The cognitive bias index is obtained by analyzing the cognitive load index over time to determine the attention distraction coefficient (such as abnormal fixation shift frequency) and the misunderstanding coefficient (such as a sudden increase in instruction response error rate).

[0191] Furthermore, the abnormal state information generation adopts a multi-level judgment logic, as shown in the following example: if the posture deviation coefficient exceeds the threshold and is accompanied by a continuous increase in the attention distraction coefficient, it is marked as "motor-cognitive joint imbalance"; if the joint impairment coefficient and misunderstanding coefficient exceed the standard simultaneously within 5 consecutive time windows, it triggers error correction instructions such as "raise the left arm 10cm", and outputs abnormal state markers and movement correction parameters to achieve closed-loop feedback control of the rehabilitation process.

[0192] This embodiment significantly improves the precision and adaptability of rehabilitation assessment through a phased weight allocation and multi-source feature fusion mechanism. The weight allocation rules are dynamically adjusted based on the user's rehabilitation duration and progress. Early rehabilitation rules focus on motor feature vectors to monitor the reconstruction of basic motor function; mid-term rehabilitation rules balance motor and cognitive feature vectors to capture synergistic anomalies during dual-task execution; and late-stage rehabilitation rules strengthen cognitive feature vectors and joint assessment indicators to evaluate higher-order cognitive recovery in complex environments. Dynamic time warping aligns motor data (acceleration variance time series) and cognitive data (response error rate time series), eliminating the temporal asynchrony of cross-modal data acquisition. Z-score standardization unifies the dimensional differences between joint range of motion and cognitive load index, ensuring the numerical comparability of the weighted fusion. The fused feature vectors are combined with the motion complexity curve to calculate the posture deviation coefficient, and the attention distraction coefficient is identified based on the cognitive load index time series. Multi-level judgment logic jointly analyzes the motor deviation index and the cognitive deviation index, enabling real-time identification and closed-loop intervention of abnormal states, effectively improving the safety and personalized adaptability of rehabilitation training.

[0193] In some embodiments, inputting abnormal state information into a dynamic threshold model to detect abnormal user behavior and generating real-time feedback instructions through preset feedback rules includes:

[0194] The initial thresholds of the dynamic threshold model are initialized based on rehabilitation information. The initial thresholds include motor deviation thresholds and cognitive deviation thresholds.

[0195] Based on the user's real-time motion information and real-time cognitive data, dynamically adjust the motion deviation threshold and cognitive deviation threshold, including:

[0196] The motion deviation threshold is dynamically adjusted based on the changing trend of the motion complexity curve, and the cognitive deviation threshold is dynamically adjusted based on the changing trend of the cognitive load index over time.

[0197] The motion deviation index is compared with the dynamically adjusted motion deviation threshold. If the motion deviation index exceeds the motion deviation threshold, it is judged as abnormal motion behavior.

[0198] The cognitive bias index is compared with the dynamically adjusted cognitive bias threshold. If the cognitive bias index exceeds the cognitive bias threshold, it is judged as cognitive behavior abnormality.

[0199] Detection results are generated based on abnormal motor behavior and cognitive behavior.

[0200] Generate preset feedback rules based on the test results;

[0201] Preset feedback rules include motion feedback rules and cognitive feedback rules;

[0202] Motion feedback rules are generated through the following steps:

[0203] Motor adjustment instructions are generated based on the type and degree of abnormal motor behavior. These instructions include adjusting the intensity, frequency, or postural requirements of the training task.

[0204] Cognitive feedback rules are generated through the following steps:

[0205] Cognitive adjustment instructions are generated based on the type and degree of cognitive behavioral abnormalities. These instructions include adjusting the complexity, attention requirements, or decision-making speed requirements of the training task.

[0206] By combining motion adjustment commands and cognitive adjustment commands, real-time feedback commands are generated.

[0207] In this embodiment, the dynamic threshold model initializes the motor deviation threshold and cognitive deviation threshold with rehabilitation information. For example, a higher motor deviation threshold is set in the early rehabilitation stage to tolerate movement instability, while the cognitive deviation threshold is set based on the initial cognitive load index to set an upper limit for the response error rate.

[0208] During the dynamic adjustment process, the motion complexity curve updates the threshold in real time by analyzing the smoothness of the motion trajectory (such as the acceleration variance volatility): if the curve shows that the motion pattern tends to be stable, the motion deviation threshold is tightened frame by frame; the cognitive load index adjusts the threshold by monitoring the efficiency of attention allocation (such as the percentage of effective fixations per unit time). If the cognitive load index continues to rise, the cognitive deviation threshold is relaxed in sync to avoid oversensitivity triggering false alarms.

[0209] When a motion deviation index (e.g., the knee joint range of motion deviates from the standard value by 18%) is detected to exceed the dynamically adjusted motion deviation threshold (e.g., the motion deviation threshold is 15%), it is judged as abnormal motion behavior (type marked as "joint angle exceeding limit"); if a cognitive deviation index (e.g., a response delay of 2.5 seconds for 3 consecutive times) exceeds the dynamic cognitive deviation threshold (e.g., the cognitive deviation threshold is 2 seconds), it is judged as abnormal cognitive behavior (type marked as "decision delay").

[0210] In the preset feedback rules, the motion feedback rules generate instructions based on the type and severity of the abnormality: if "joint angle exceeds the limit" and the severity is moderate, a voice prompt of "Please reduce the knee flexion angle by 5 degrees" is triggered, and the training intensity is reduced (e.g., resistance is reduced by 10%). The cognitive feedback rules generate instructions for "decision delay": reduce the number of digits in the arithmetic problem from 3 digits to 2 digits, and extend the response time requirement to 3 seconds. When motion behavior abnormalities and cognitive behavior abnormalities occur simultaneously (e.g., shoulder joint elevation is less than 12cm and two consecutive instructions are misunderstood), the real-time feedback instructions will combine motion adjustment ("Raise the left arm to the 15cm mark") and cognitive intervention ("Please repeat the current training steps"), output synchronously through voice and interface highlight prompts to ensure the coordination and executability of error correction instructions.

[0211] This embodiment significantly improves the accuracy of rehabilitation training anomaly detection and the timeliness of intervention through a dynamic threshold model and adaptive feedback mechanism. The motion deviation threshold and cognitive deviation threshold, initialized based on rehabilitation information, can adapt to the needs of different rehabilitation stages. During dynamic adjustment, the motion complexity curve tightens the motion deviation threshold frame by frame through motion trajectory smoothness analysis, while the cognitive load index dynamically widens the cognitive deviation threshold according to changes in attention allocation efficiency, avoiding misjudgments caused by user fatigue or task overload. When the motion deviation index exceeds the dynamic motion deviation threshold, a motion feedback rule is triggered, generating voice prompts and adjusting the intensity, frequency, or posture requirements of the training task. If the cognitive deviation index exceeds the cognitive deviation threshold, a cognitive feedback rule is activated, adjusting the complexity, attention requirements, or decision-making speed requirements of the training task. For scenarios where motion and cognitive behavioral anomalies occur concurrently, feedback instructions combine motion and cognitive adjustment instructions to achieve coordinated execution of multimodal error correction instructions, forming a real-time control chain from anomaly detection to closed-loop correction, effectively ensuring training safety and personalized adaptation capabilities.

[0212] Please see Figure 4 In a second aspect, this embodiment also provides a motor and cognitive ability assessment system, applicable to the assessment method described in the first aspect, the system comprising:

[0213] The information acquisition module 111 is used to acquire the user's real-time motion information. The real-time motion information includes real-time motion data and real-time cognitive data. The real-time motion data includes real-time joint angle information and real-time gait information. The real-time cognitive data includes real-time attention information and real-time decision speed information.

[0214] The logic processing module 112 is used to construct a physiological-behavioral coupling model, which is configured to be generated based on a hidden Markov model. Real-time motion information is input into the physiological-behavioral coupling model to obtain a joint evaluation index, which includes a motion complexity curve and a cognitive load index time series. A hierarchical feature extraction model is constructed, which includes a feature extraction unit and a feature fusion unit. Real-time motion information is input into the feature extraction unit to obtain motion feature vectors and cognitive feature vectors. The motion feature vectors, cognitive feature vectors, and joint evaluation index are input into the feature fusion unit to obtain abnormal state information, which includes motion abnormalities and cognitive abnormalities.

[0215] The anomaly feedback module 113 is used to input abnormal status information into the dynamic threshold model to detect abnormal user behavior and generate real-time feedback instructions through preset feedback rules.

[0216] The instruction sending unit 114 is used to adjust the user's training task parameters in real time based on real-time feedback instructions.

[0217] The method steps mentioned in the above system are consistent with the method steps described in the first aspect, and will not be repeated here in this embodiment.

[0218] Furthermore, in some preferred embodiments, the evaluation system 11 also includes a miniaturized near-infrared brain imaging module, employing fNIRS (functional near-infrared spectroscopy) technology to non-invasively acquire prefrontal cortex blood oxygenation signals via a head-worn device, generating real-time blood oxygenation concentration time-series data to quantify cognitive load. This module is periodically activated during training tasks (e.g., detecting for 30 seconds every 10 minutes), identifying cognitive resource consumption levels by detecting changes in hemoglobin concentration (e.g., an increase of 0.05 mmol / L in deoxyhemoglobin concentration), and synchronizing the data to the information acquisition module 111 as supplementary input to real-time cognitive data (expanding existing real-time attention and decision-making speed information). When calculating the cognitive load index time-series, the physiological-behavioral coupling model of the logic processing module 112 integrates blood oxygenation concentration time-series data (e.g., the area of ​​the activated prefrontal cortex exceeds...). The system uses time markers to indicate high-load conditions, combined with real-time decision-making speed information (such as response delays exceeding 1.8 seconds), to generate more accurate criteria for determining cognitive abnormalities. The miniaturized design ensures the device is lightweight and does not restrict head movement freedom when worn. Its periodic detection data is also used to calibrate cognitive deviation thresholds; for example, when the blood oxygen recovery rate is 20% lower than the baseline, the dynamic threshold model relaxes the cognitive deviation threshold by 5% to adapt to the user's current neurometabolic state.

[0219] The motor and cognitive ability assessment system 11 provided in this embodiment achieves comprehensive and precise monitoring and personalized adaptation of the rehabilitation training process through multi-dimensional data fusion and dynamic closed-loop control mechanisms. The information acquisition module 111 synchronously collects real-time joint angle information, real-time gait information, real-time attention information, and real-time decision-making speed information to construct a complete motor and cognitive behavior dataset. In the logic processing module 112, a physiological-behavioral coupling model based on a Hidden Markov Model maps the above data into a motor complexity curve and a cognitive load index time series. A hierarchical feature extraction model generates motor feature vectors and cognitive feature vectors through a feature extraction unit, which are then associated with joint evaluation indicators through a feature fusion unit to accurately identify motor abnormalities or cognitive abnormalities. The abnormality feedback module 113 uses a dynamic threshold model to determine behavioral deviations in real time and generates targeted instructions through preset feedback rules. Furthermore, the miniaturized near-infrared brain imaging module periodically detects blood oxygenation signals in the prefrontal cortex using fNIRS technology, capturing changes in deoxyhemoglobin concentration to quantify cognitive resource consumption. This data is synchronized to the information acquisition module 111 to supplement real-time cognitive data, enabling the physiological-behavioral coupling model to integrate blood oxygenation time series and decision-making speed when calculating the cognitive load index, significantly improving the neurophysiological basis for determining cognitive abnormalities. The miniaturized design ensures unimpeded head movement freedom, while periodic blood oxygenation data dynamically calibrates the cognitive deviation threshold, forming a closed-loop adaptation from neurometabolic state to behavioral performance. Finally, the command sending unit 114 optimizes training parameters in real time, achieving the goal of coordinated rehabilitation of motor ability and cognitive function.

[0220] Please see Figure 5 In a third aspect, this embodiment also provides a motor and cognitive ability assessment device, including a smart terminal 12, a motion acquisition device 13, a cognitive acquisition device 14, an audio playback device 15, and an auxiliary correction device 16. The smart terminal 12 is equipped with the assessment system 11 described in the second aspect; the motion acquisition device 13 is electrically connected to the smart terminal 12 and is used to acquire real-time motion data; the cognitive acquisition device 14 is electrically connected to the smart terminal 12 and is used to acquire real-time cognitive data; the audio playback device 15 is electrically connected to the smart terminal 12 and is used to provide audio guidance to the user based on real-time feedback instructions from the smart terminal 12; the auxiliary correction device 16 is electrically connected to the smart terminal 12 and is used to correct the user's movements based on real-time feedback instructions so that the user's real-time movements meet the requirements of the training task.

[0221] The motor and cognitive ability assessment device 1 described in this embodiment achieves synchronous quantitative assessment and training adaptation of motor and cognitive abilities in a home environment through multimodal interaction and adaptive feedback mechanisms. The motion acquisition device 13 can use wearable inertial sensors (such as a six-axis sensor built into a knee joint strap) to capture real-time joint angle information and gait parameters, and the data is transmitted to the smart terminal 12 via Bluetooth Low Energy.

[0222] The cognitive acquisition device 14 can be a smart glasses with an integrated eye-tracking module. It records attention allocation characteristics by capturing eye movement trajectories and, in conjunction with a handheld response button device, simultaneously collects decision response time and records real-time attention information and decision speed information when completing digital connection tests.

[0223] The assistive correction device 16 can employ a dynamic knee joint limiting brace with a built-in angle sensor to monitor the flexion and extension range in real time. When the logic processing module 112 detects abnormal movement, it dynamically adjusts the limiting angle via a stepper motor to force the user's movements to conform to training specifications. In addition, the assistive correction device 16 can also be an adjustable wrist resistance brace, which dynamically adjusts the rotational damping via a micro servo motor to force the user to maintain a standard grip posture.

[0224] Furthermore, the evaluation device 1 provided in this embodiment may also include an auxiliary feedback device, which may include a multi-degree-of-freedom mechanical feedback device and a tactile navigation belt.

[0225] The multi-degree-of-freedom mechanical feedback device includes a three-axis linkage robotic arm with a built-in torque sensor to simulate the resistance of everyday activities (such as simulating the action of lifting a kettle to pour water). (Resistance curve) is used to test a user's ability to adjust strategies in response to sudden changes in load during home training, for example, when resistance increases stepwise. Joint torque compensation response time Second.

[0226] The haptic navigation belt is equipped with ring-shaped vibration motors that encode directional commands through vibration patterns (e.g., two consecutive vibrations to the right represent a 30° right turn). When the smart terminal 12 generates a virtual obstacle scene (e.g., random path deviation prompts), it assesses the user's cognitive-motor coordination ability (e.g., directional correction delay). Second).

[0227] Optionally, an odor release module is also included as a backup component. The odor release module uses a sealed atomization chamber to store standard odor preparations (such as menthol) and releases them only when the attention shift assessment mode is activated to test the stability of motor control when the user's attention is shifted.

[0228] Preferably, the mechanical feedback device features a foldable storage design, the tactile belt weighs only 180g and is covered with a non-slip silicone layer, and the odor module allows for quick replacement of the preparation chamber via a magnetic interface. All devices are powered via a unified Type-C interface, ensuring that elderly users can assemble, charge, and perform basic maintenance without assistance, achieving safe and reliable motor-cognitive integrated rehabilitation training through natural human-computer interaction.

[0229] The motor and cognitive ability assessment device 1 described in this embodiment captures joint angles and gait parameters in real time through the motion acquisition device 13, and simultaneously records attention and decision-making speed information through the cognitive acquisition device 14, forming a multi-dimensional behavioral data chain. The auxiliary correction device 16 dynamically adjusts the limiting angle of the rigid support based on the anomaly detection results of the logic processing module 112 to ensure standardized movements. The multi-degree-of-freedom mechanical feedback device simulates daily resistance environments through a three-axis linkage robotic arm, accurately testing the user's ability to adjust strategies to cope with sudden load changes; the tactile navigation belt uses a ring vibration motor to encode directional commands, quantifying cognitive-motor coordination efficiency in a virtual obstacle scenario. The odor release module releases standard odor stimuli through a sealed atomization chamber to assess motor stability during attention shifts. All components adopt a foldable storage design, lightweight construction, and magnetic dosage chamber, along with a unified Type-C power supply interface, reducing operational complexity and allowing elderly users to independently assemble and maintain the device. This achieves a closed-loop adaptation of motor function enhancement and cognitive ability improvement in a home environment, ensuring the safety and effectiveness of rehabilitation training.

[0230] In some embodiments, the smart terminal 12 is configured as at least one of a tablet, computer, or mobile phone; the motion acquisition device 13 is configured as at least one of a camera, camcorder, sensor mat, radar, or infrared sensor; and the cognitive acquisition device 14 is configured as at least one of a smart bracelet, smart helmet, smart glasses, headphones, camera, camcorder, voice interaction device, or VR device.

[0231] In this embodiment, the motion acquisition device 13 uses a flexible piezoresistive material layered structure for zoned pressure detection. Pressure sensing units are distributed on the surface of the sensing mat. The pressure distribution of the foot is captured in real time by the change of the piezoresistive value, supporting dynamic center of gravity tracking and gait phase recognition.

[0232] Motion capture device 13 covers multi-dimensional motion capture scenarios through a combination of cameras, video cameras, sensor mats, radar detectors, and infrared sensors. The sensor mats employ a complementary design combining a pressure-sensitive capacitive matrix floor and a flexible piezoresistive yoga mat, achieving precise analysis from macroscopic motion trajectories to microscopic pressure characteristics. Cognitive capture device 14, through the coordinated operation of smart bracelets, smart helmets, smart glasses, headphones, voice interaction devices, and VR devices, simultaneously acquires user physiological data (such as heart rate and EEG signals), visual focus trajectories, voice commands, and virtual environment interaction behaviors, constructing a multi-dimensional cognitive state map. The deep integration of motion and cognitive data enables smart terminal 12 (tablet, computer, or mobile phone) to generate personalized training programs. For example, when the sensor mat detects gait imbalance, it adjusts the AR projection guidance intensity based on visual attention data from the smart glasses; when abnormal pressure distribution triggers an alarm, it broadcasts corrective instructions in real time through the voice interaction device. This embodiment ensures the synchronization accuracy of motion capture and the real-time response of feedback through protocol compatibility between devices and data spatiotemporal alignment mechanisms. At the same time, it relies on a modular architecture to support flexible configuration for different rehabilitation stages or training scenarios, thereby improving the adaptability and systematic nature of home rehabilitation training while ensuring user privacy and data security.

[0233] In some embodiments, the sensing mat includes a pressure-sensitive capacitive matrix floor and a flexible piezoresistive material yoga mat. The pressure-sensitive capacitive matrix floor is used to capture the user's gait distribution and center of gravity trajectory in real time, and the flexible piezoresistive material yoga mat is used to detect the user's pressure distribution in zones.

[0234] In this embodiment, the zoned pressure detection of the flexible piezoresistive material yoga mat is achieved through the following steps:

[0235] A multilayer flexible piezoresistive film is embedded in the base of the yoga mat. Each layer of film has an independent detection area formed by microcircuit etching, dividing the mat surface into square grid units. A parallel array of piezoresistive resistors is set in each unit.

[0236] When the user applies pressure, the piezoresistive material in the pressure area deforms, causing a change in the resistance of the piezoresistor in the corresponding grid cell. The change in resistance is converted into a voltage signal through a constant current source circuit.

[0237] The signal acquisition module polls and scans each grid cell at a sampling frequency of 100Hz, quantizes the voltage signal using an analog-to-digital converter, and calculates the pressure value of each zone in real time.

[0238] The data processing module performs spatial clustering on adjacent grid cells, identifies the boundaries of continuous pressure areas, and, combined with the preset human foot / hand geometric template, distinguishes the left / right foot or upper limb support areas.

[0239] The pressure distribution weight of each zone (such as the pressure percentage of the right leg support area) is calculated using the pressure center algorithm. (It triggers an early warning in real time) and aligns the partition pressure data with the spatial coordinates of the AR projection marker in time and space to generate dynamic feedback instructions.

[0240] The collaborative work of the sensor mat and AR projection markers can be understood as follows: when a user walks along a virtual straight line, the piezoresistive array detects the heel-forefoot pressure transfer trajectory in real time, triggering dynamic adjustments to the AR path and further enhancing the training guidance effectiveness.

[0241] In this embodiment, the pressure-sensitive capacitive matrix floor captures the user's gait distribution and center of gravity trajectory in real time through changes in capacitance, forming macroscopic motion trajectory data. The flexible piezoresistive yoga mat embeds multiple layers of flexible piezoresistive films within its base and uses microcircuit etching technology to divide the mat surface into square grid units. Each unit contains a parallel array of piezoresistive resistors. When the user applies pressure, the deformation of the piezoresistive material in the pressure area causes a change in the resistance of the corresponding grid. A constant current source circuit converts the resistance change into a voltage signal. The signal acquisition module polls and scans each grid unit at a sampling frequency of 100Hz, and after quantization by an analog-to-digital converter, calculates the pressure value of each zone in real time. The data processing module performs spatial clustering on adjacent grids, combines a preset human foot / hand geometric template to identify the boundaries of continuous pressure areas, accurately distinguishes the left / right foot or upper limb support areas, and calculates the pressure distribution weight of each zone using a pressure center algorithm. As the user walks along a virtual straight line, the piezoresistive array of the flexible piezoresistive yoga mat detects the pressure transfer trajectory from the heel to the forefoot in real time, triggering dynamic adjustments to the AR path. This causes the virtual projected markers to shift or enhance visual cues based on the actual pressure distribution, thereby optimizing the real-time performance and accuracy of training guidance. This embodiment achieves cross-scale data fusion from global motion trajectory to local pressure distribution through the complementary functions of the pressure-sensitive capacitive matrix floor and the flexible piezoresistive yoga mat. Combined with the dynamic feedback of AR projection, it forms a closed-loop training system, effectively improving the efficiency of gait correction and balance training.

[0242] Furthermore, the following examples can be derived from the above technical solutions:

[0243] The user stands on the pressure pad, faces the tablet, and starts the "Balance Training Mode";

[0244] An AR virtual coach demonstrates a squatting exercise, with motion acquisition device 13 capturing the user's joint points and cognitive acquisition device 14 monitoring acceleration.

[0245] If knee valgus is detected, the cognitive acquisition device 14 will prompt "both feet externally rotated 15°";

[0246] Cognitive task overlay: While squatting, a tablet displays random numbers, and the user needs to read aloud the second-to-last number;

[0247] Response delay If the number of seconds is specified, the number of digits in the subsequent numbers will be automatically reduced (e.g., from 5 digits to 3 digits).

[0248] Training complete and report generated:

[0249] Motor score: the percentage of joint range of motion that meets the target (e.g., 85% of the joints can flex to 90°).

[0250] Cognitive assessment: average response time (1.2 seconds / question), error rate (5%).

[0251] Unlike existing technologies, the above technical solution has the following beneficial effects:

[0252] This invention achieves coordinated dynamic assessment of motor and cognitive abilities by constructing a physiological-behavioral coupling model and a hierarchical feature extraction model. Through the joint assessment indicators (motor complexity curve and cognitive load index time series) generated by the Hidden Markov Model, combined with time-domain, frequency-domain, and spatiotemporal multi-dimensional feature extraction, it can accurately capture the coupling anomalies between motor compensatory behaviors and cognitive load imbalances. Based on dynamically adjusted weight allocation rules and an adaptive threshold detection mechanism for different rehabilitation stages, it can adapt to the personalized needs of different rehabilitation cycles (early, middle, and late stages), identify abnormal states such as postural deviations, joint disorders, and attention deficits in real time, and adjust training intensity, complexity, and decision-making speed requirements through dynamic feedback rules, effectively avoiding secondary injuries caused by the accumulation of incorrect movements. Simultaneously, the joint probability distribution constructed by the kernel density estimation method and the observation space of the Hidden Markov Model can analyze the temporal correlation of physiological and behavioral patterns without the need for complex sensors, breaking through the dependence of traditional assessments on professional equipment. This provides real-time movement correction and cognitive training optimization capabilities for home rehabilitation, significantly improving the safety and effectiveness of self-training.

[0253] Finally, it should be noted that although the above embodiments have been described in the text and drawings of this application, this should not limit the scope of patent protection of this application. Any technical solutions that are based on the essential concept of this application and utilize the content described in the text and drawings of this application, resulting in equivalent structural or procedural substitutions or modifications, as well as the direct or indirect application of the technical solutions of the above embodiments to other related technical fields, are all included within the scope of patent protection of this application.

Claims

1. A method for assessing motor and cognitive abilities, characterized in that, The method includes: The system acquires the user's real-time motion information, which includes real-time motion data and real-time cognitive data. The real-time motion data includes real-time joint angle information and real-time gait information, and the real-time cognitive data includes real-time attention information and real-time decision speed information. A physiological-behavioral coupling model is constructed, which is configured to be generated based on a hidden Markov model. The real-time motion information is input into the physiological-behavioral coupling model to obtain a joint evaluation index, which includes a motion complexity curve and a cognitive load index time series. A hierarchical feature extraction model is constructed, which includes a feature extraction unit and a feature fusion unit. The real-time motion information is input into the feature extraction unit to obtain motion feature vectors and cognitive feature vectors. The motion feature vector, cognitive feature vector, and joint evaluation index are input into the feature fusion unit to obtain abnormal state information, which includes motion abnormalities and cognitive abnormalities. The abnormal status information is input into the dynamic threshold model to detect abnormal user behavior, and real-time feedback instructions are generated through preset feedback rules. The user's training task parameters are adjusted in real time according to the real-time feedback instructions. Construct a physiological-behavioral coupling model, which is configured to be generated based on a hidden Markov model, including: Acquire sample motion information, which includes sample motion data and sample cognitive data; The sample motion information is processed by kernel density estimation to obtain joint probability distribution information, which includes the user's physiological-behavioral pattern at a specific time point. The joint probability distribution information is mapped to the hidden space of the Hidden Markov Model; In addition, the length of the first preset time window is defined, and the sample motion information is divided into multiple first preset time windows in chronological order; The motion information of the samples in each first preset time window is time-aligned and normalized to obtain multiple joint observation vectors; Multiple joint observation vectors are arranged in chronological order and a joint observation sequence is generated to form the observation space of the hidden Markov model. Based on the hidden space and the observation space, the transition probability matrix and observation probability matrix of the Hidden Markov Model are constructed and trained to obtain the trained physiological-behavioral coupling model. The motion complexity curve is configured to be generated by calculating the transition frequency of the hidden space and the entropy value of the joint observation sequence, and the cognitive load index time series is configured to be generated by calculating the duration of the hidden space and the variance of the joint observation sequence.

2. The method for assessing motor and cognitive abilities as described in claim 1, characterized in that, The motion feature vector is configured to be obtained through the following steps: The real-time motion data is divided according to a second preset time window to obtain multiple first time series segments, and each time series segment corresponds to real-time motion data within a second preset time window; For each first time series segment, a first time-domain feature is extracted, which includes a first mean, a first variance, a first maximum value, a first minimum value, and a first root mean square value. For each first time series segment, the time-domain signal is converted into a frequency-domain signal by a fast Fourier transform, and the first frequency-domain features are extracted. The first frequency-domain features include the first dominant frequency, the first spectral energy, and the first spectral entropy. For each first time series segment, extract the first spatiotemporal features, which include the spatial distribution of the motion trajectory, the spatial variation of the motion velocity, and the spatial variation of the motion acceleration. The motion feature vector is generated by combining the first time-domain feature, the first frequency-domain feature, and the first spatiotemporal feature.

3. The method for assessing motor and cognitive abilities as described in claim 1, characterized in that, The cognitive feature vector is configured to be obtained through the following steps: The real-time cognitive data is divided into a third preset time window to obtain multiple second time series segments, and each second time series segment corresponds to real-time cognitive data within a third preset time window. For each second time series segment, extract the second time-domain feature, which includes the second mean, the second variance, the second maximum value, and the second minimum value. For each second time series segment, the time-domain signal is converted into a frequency-domain signal by a fast Fourier transform, and the second frequency-domain features are extracted. The second frequency-domain features include the second dominant frequency, the second spectral energy, and the second spectral entropy. For each second time series segment, a second spatiotemporal feature is extracted, which includes spatial changes in attention distribution, spatial changes in decision speed, and spatial changes in cognitive load, to characterize signs of cognitive decline in the user in the spatial dimension. The second time-domain feature, the second frequency-domain feature, and the second spatiotemporal feature are combined to generate the cognitive feature vector.

4. The method for assessing motor and cognitive abilities as described in claim 1, characterized in that, The motion feature vector, cognitive feature vector, and joint evaluation index are input into the feature fusion unit to obtain abnormal state information, including: The user's rehabilitation information is obtained, including rehabilitation duration and rehabilitation progress. Based on the rehabilitation information, the weight allocation rules of the motor feature vector, cognitive feature vector and joint evaluation index are initialized. The weight allocation rules are configured as early rehabilitation rules, mid-term rehabilitation rules and late rehabilitation rules. The motion feature vector, cognitive feature vector, and joint evaluation index are subjected to feature alignment and normalization. The motion feature vector, cognitive feature vector, and joint evaluation index after feature alignment and normalization are weighted and fused to obtain a fused feature vector. The fused feature vector is used to calculate the motion deviation index based on the motion complexity curve. The motion deviation index includes the posture deviation coefficient, joint impairment coefficient, or muscle contraction impairment coefficient. The fused feature vector is used to calculate the cognitive bias index based on the cognitive load index time series. The cognitive bias index includes the user's sluggishness coefficient, misunderstanding coefficient, and attention distraction coefficient during the cognitive process. Abnormal state information is generated based on the motion deviation index and the cognitive deviation index.

5. The method for assessing motor and cognitive abilities as described in claim 4, characterized in that, The abnormal state information is input into a dynamic threshold model to detect abnormal user behavior, and real-time feedback instructions are generated according to preset feedback rules, including: The initial thresholds of the dynamic threshold model are initialized based on the rehabilitation information, and the initial thresholds include a motor deviation threshold and a cognitive deviation threshold. Based on the user's real-time motion information and real-time cognitive data, the motion deviation threshold and cognitive deviation threshold are dynamically adjusted, including: The motion deviation threshold is dynamically adjusted according to the changing trend of the motion complexity curve, and the cognitive deviation threshold is dynamically adjusted according to the changing trend of the cognitive load index over time. The motion deviation index is compared with the dynamically adjusted motion deviation threshold. If the motion deviation index exceeds the motion deviation threshold, the motion behavior is determined to be abnormal. The cognitive bias index is compared with the dynamically adjusted cognitive bias threshold. If the cognitive bias index exceeds the cognitive bias threshold, it is determined to be a cognitive behavior abnormality. Detection results are generated based on the described abnormal motor and cognitive behaviors. Generate preset feedback rules based on the detection results; The preset feedback rules include motion feedback rules and cognitive feedback rules; The motion feedback rules are generated through the following steps: Based on the type and degree of the abnormal motor behavior, a motor adjustment instruction is generated, which includes adjusting the intensity, frequency, or posture requirements of the training task. The cognitive feedback rules are generated through the following steps: Cognitive adjustment instructions are generated based on the type and degree of the cognitive behavioral abnormalities. These instructions include adjusting the complexity, attention requirements, or decision-making speed requirements of the training task. The motion adjustment command and the cognitive adjustment command are combined to generate the real-time feedback command.

6. A system for assessing motor and cognitive abilities, characterized in that, The system, applicable to the evaluation method according to any one of claims 1 to 5, comprises: The information acquisition module is used to acquire the user's real-time motion information, which includes real-time motion data and real-time cognitive data. The real-time motion data includes real-time joint angle information and real-time gait information, and the real-time cognitive data includes real-time attention information and real-time decision speed information. A logic processing module is used to construct a physiological-behavioral coupling model, which is configured to be generated based on a hidden Markov model. The real-time motion information is input into the physiological-behavioral coupling model to obtain a joint evaluation index, which includes a motion complexity curve and a cognitive load index time series. A hierarchical feature extraction model is also constructed, comprising a feature extraction unit and a feature fusion unit. The real-time motion information is input into the feature extraction unit to obtain motion feature vectors and cognitive feature vectors. The motion feature vectors, cognitive feature vectors, and the joint evaluation index are input into the feature fusion unit to obtain abnormal state information, which includes motion abnormalities and cognitive abnormalities. An anomaly feedback module is used to input the anomaly status information into a dynamic threshold model to detect abnormal user behavior and generate real-time feedback instructions through preset feedback rules. The instruction sending unit is used to adjust the user's training task parameters in real time according to the real-time feedback instructions.

7. A device for assessing motor and cognitive abilities, characterized in that, include: A smart terminal, on which the evaluation system of claim 6 is mounted; A motion acquisition device is electrically connected to the smart terminal, and the motion acquisition device is used to acquire the real-time motion data; A cognitive acquisition device is electrically connected to the smart terminal, and the cognitive acquisition device is used to acquire the real-time cognitive data; An audio playback device is electrically connected to the smart terminal, and the audio playback device is used to provide audio guidance to the user based on real-time feedback instructions from the smart terminal. An auxiliary correction device is electrically connected to the smart terminal. The auxiliary correction device is used to correct the user's movements according to the real-time feedback instructions so that the user's real-time movements meet the requirements of the training task.

8. The motor and cognitive ability assessment device as described in claim 7, characterized in that, The smart terminal is configured as at least one of a tablet, a computer, and a mobile phone; The motion capture device is configured as at least one of a camera, video camera, sensor mat, radar, and infrared sensor; The cognitive acquisition device is configured as at least one of a smart bracelet, smart helmet, smart glasses, headphones, camera, camcorder, voice interaction device, and VR device.

9. The motor and cognitive ability assessment device as described in claim 8, characterized in that, The inductive mat includes a pressure-sensitive capacitive matrix floor and a flexible piezoresistive material yoga mat. The pressure-sensitive capacitive matrix floor is used to capture the user's gait distribution and center of gravity trajectory in real time, and the flexible piezoresistive material yoga mat is used to detect the user's pressure distribution in zones.

Images