Exercise coordination non-contact screening system based on computer vision

Abstract

The invention discloses a non-contact screening system for motion coordination based on computer vision, which relates to the field of computer vision, and comprises the following steps: acquiring a human body motion image through a non-contact camera, extracting human body key point coordinates frame by frame by using a deep learning algorithm after noise reduction and distortion correction, and constructing a time sequence data set; performing smooth jitter removal and interpolation processing to form standardized data; calculating track, rhythm and stability multi-dimensional motion coordination quantitative characteristics based on the data, and performing comparative analysis with a personal historical database to identify a motion function degradation trend; and based on age groups and scenes, customizing screening parameters, coordinating workflows of the components, and realizing full-process automatic screening. The method has the advantages that invasive and cross infection risks are avoided through non-contact acquisition, accurate and objective assessment of the exercise coordination is realized by means of deep learning and time series data analysis, exercise function degradation can be early warned, and an efficient and reliable scheme is provided for various exercise coordination screening scenes.

Description

Technical Field

[0001] This invention relates to the field of computer vision, and in particular to a non-contact screening system for motion coordination based on computer vision. Background Technology

[0002] Non-contact screening systems for motor coordination are intelligent health assessment tools developed in recent years based on computer vision and artificial intelligence technologies. They aim to solve the problems of low efficiency, high subjectivity, and reliance on specialized equipment in traditional motor function testing. The demand for early screening of motor dysfunction caused by neurodegenerative diseases has surged, but current clinical assessments mostly rely on scale observation or wearable sensors, which are limited by high cost and operational complexity.

[0003] Most current motor coordination screening systems on the market still rely on contact-based data collection devices (such as wearable sensors and force plates), requiring subjects to actively cooperate with the wearing, resulting in poor compliance and making them unsuitable for people with mobility impairments. At the algorithm level, they largely depend on traditional computer vision technology, which suffers from insufficient accuracy in extracting key human body points and limited quantification features, making it difficult to comprehensively assess motor coordination. Furthermore, most systems lack the ability to manage personal historical data and perform time-series comparisons, failing to provide early warnings of motor function decline trends. Their parameter configurations are also fixed, making it difficult to adapt to different age groups and screening scenarios. Some systems also suffer from inadequate data encryption mechanisms and poor operational stability, resulting in limited overall screening efficiency and reliability. Summary of the Invention

[0004] To improve the existing system, a computer vision-based contactless screening system for motor coordination is provided. This system avoids the risks of intrusion and cross-infection by collecting data without contact. It relies on deep learning and time-series data analysis to achieve accurate and objective assessment of motor coordination, can provide early warning of motor function degradation, supports personalized configuration for multiple scenarios, and is stable and secure in operation, providing an efficient and reliable solution for various motor coordination screening scenarios.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A computer vision-based non-contact screening system for motion coordination includes:

[0007] Non-contact image acquisition module: Captures a continuous sequence of motion images of the person being screened using a non-contact camera, and performs noise reduction and distortion correction on the acquired images;

[0008] Human body key point extraction module: The acquired motion image sequence is analyzed frame by frame using a deep learning-based computer vision algorithm to extract the coordinates of the human body key nodes of the screened person in each frame, forming a set of human body key points corresponding to each frame.

[0009] Time series data processing module: It associates and integrates the set of human key points in chronological order to generate a human key point time series dataset. After smoothing and de-jittering, it fills in the missing data through time series interpolation to form standardized human key point time series data.

[0010] Quantitative Feature Calculation Module: Based on standardized human key point time series data, it calculates multi-dimensional quantitative feature indicators that characterize motion coordination, including trajectory features, rhythm and speed features, stability features, coordination features, and spatiotemporal features.

[0011] Personal historical data management module: Stores quantitative characteristic index data of the screened individuals, establishes unique personal exercise history data for each screened individual, and classifies and manages data based on the screened individual's identity identifier;

[0012] Functional decline trend module: The quantitative characteristic indicators of this screening are compared with the individual's exercise history data over time to obtain the changing trend of each quantitative characteristic indicator and determine whether there is a trend of motor function decline.

[0013] Screening parameter configuration module: used to preset and dynamically adjust the core parameters in the screening process, and to customize parameter combinations according to different age groups and different screening scenarios;

[0014] System control module: Establishes communication connections with each module, has built-in module scheduling algorithms and timing control logic, coordinates the working sequence of each module, controls the start, pause, termination and restart of the screening process, and realizes real-time data transmission and interactive verification between modules.

[0015] Preferably, the contactless image acquisition module specifically includes:

[0016] Camera unit: Employs a depth camera, positioned directly in front of the screening area, to capture a continuous sequence of motion images of the person being screened in real time;

[0017] Image preprocessing unit: Built-in dedicated image processing chip, which uses adaptive noise reduction algorithm to specifically eliminate environmental noise and electronic noise, and performs geometric distortion correction based on preset lens calibration parameters;

[0018] Dynamic tracking unit: Automatically optimizes the detection accuracy of key points in occluded areas.

[0019] Preferably, the human body key point extraction module specifically includes:

[0020] Image sequence receiving unit: Connects to the contactless image acquisition module to receive the processed motion image sequence of the screened person in real time, and uses a first-in-first-out buffering mechanism for caching;

[0021] Deep learning parsing unit: Based on the deep learning model in Azure Kinect Body Tracking SDK, it parses the image sequence frame by frame and extracts the three-dimensional coordinates of 32 key points of the whole body, including face, posture and hands;

[0022] Key point verification and optimization unit: It determines the occlusion status by the correlation of key point positions in adjacent frames, removes invalid coordinates caused by occlusion, optimizes key point coordinates by using a temporal smoothing filtering algorithm, corrects random errors, and forms a set of human body key points corresponding to each frame.

[0023] Preferably, the time series data processing module specifically includes:

[0024] Temporal correlation and integration unit: Based on the timestamp synchronization algorithm, the coordinates of key human body points frame by frame are correlated and bound according to the acquisition time sequence, generating a structured time series dataset containing the three-dimensional spatial coordinates and visibility parameters of each key point and timestamp;

[0025] Data optimization unit: Adaptive temporal de-jitter algorithm is used to eliminate random noise in key point coordinates. For missing data caused by occlusion, temporal interpolation is performed to supplement it based on effective data from previous and next frames.

[0026] Data output adaptation unit: Converts the optimized standardized time series dataset into a format compatible with the quantization feature calculation module and transmits it in real time.

[0027] Preferably, the quantization feature calculation module specifically includes:

[0028] Trajectory feature unit: The smoothness of the motion path is obtained by calculating the variance of the curvature change of the path curve; the accuracy of the motion execution is measured by the deviation value between the actual motion path and the standard motion path; and the presence of tremor during the motion is judged by the instantaneous displacement fluctuation amplitude of key points.

[0029] Rhythm and speed feature unit: The movement cycle is calculated based on the time interval between two consecutive completions of the same type of movement, and the speed symmetry and acceleration symmetry of the corresponding movements of the left and right limbs are calculated by the ratio of the peak speed and acceleration of the corresponding key points of the bilateral limbs.

[0030] Stability feature unit: Fit the center of gravity position by the coordinates of key points of the hip and torso, calculate the displacement range of the center of gravity in three-dimensional space, and measure the posture stability by the change in the center of gravity offset and the body tilt angle during single-leg support.

[0031] Coordination feature unit: The synchronicity of the left and right limb movements is assessed by the time difference of the key points of the bilateral limbs reaching the extreme position of the movement, and the symmetry of the left and right limb movements is calculated by the similarity of the movement amplitude and movement trajectory of the corresponding limbs on both sides.

[0032] Spatiotemporal characteristic unit: including the total time to complete the preset specified action, the number of times the action is completed per unit time, the time proportion of each action stage, and the connection time of action transition.

[0033] Preferably, the personal historical data management module specifically includes:

[0034] Tiered storage unit: Equipped with a hybrid storage architecture consisting of SSD arrays and mechanical hard drives. The SSD array stores the screening data that has been accessed frequently in the past 3 years, while the mechanical hard drives archive long-term historical data, and independent data partitions are created according to the identity of the screened person.

[0035] Encryption and Security Unit: Integrates an AES-256 encryption module to encrypt stored data throughout the process and implements access control based on role-based access permissions;

[0036] Data interaction and scheduling unit: Supports querying historical data by identity and time range, and can link with the function degradation trend module to push historical data for a specified time period.

[0037] Preferably, the functional degradation trend module specifically includes:

[0038] Time-series comparative analysis unit: This unit performs time-series comparative analysis between the quantitative characteristic indicators obtained from this screening and the historical indicator data in the individual's historical data archive.

[0039] Trend identification unit: By calculating the rate of change, fluctuation range and trend fit of each quantitative characteristic indicator within a continuous screening period, a personal motor function change curve is constructed, and the presence of a motor function degeneration trend is identified based on the curve;

[0040] Early warning unit: When the change range of one or more quantitative characteristic indicators exceeds the preset normal fluctuation threshold, or the change trend shows a continuous decline, a functional degradation early warning signal is generated. The early warning signal includes the abnormal characteristic indicator type, degradation degree assessment and change trend.

[0041] Data output unit: Outputs trend analysis results and change curve data, and synchronously saves analysis logs to personal exercise history data.

[0042] Preferably, the screening parameter configuration module specifically includes:

[0043] Hierarchical parameter management unit: Based on the hierarchical storage of parameters, the data acquisition class includes resolution and frame rate, the feature calculation class includes smoothness threshold, and the early warning class includes degradation fluctuation threshold.

[0044] Scenario-based parameter adaptation unit: Built-in multi-scenario preset parameter templates, including community health screening, medical institution special screening, and elderly health monitoring. The templates optimize parameter combinations for different age groups.

[0045] Parameter synchronization and verification unit: pushes configuration parameters to each module in real time through a clock synchronization mechanism, and verifies the module's operating status after the parameters are applied.

[0046] Preferably, the system control module specifically includes:

[0047] Timing scheduling unit: Based on the phased development of the screening process, a module coordination timing table is formulated. The working rhythm of each module is coordinated through clock synchronization signals. By adjusting the timing matching relationship between contactless image acquisition and key point extraction, a time-coherent data transmission link is obtained.

[0048] Data interaction verification unit: It establishes a bidirectional communication link with each module through a multi-channel data interface matrix, uses the CRC check algorithm to verify cross-module transmitted data, and triggers a retransmission mechanism for abnormal data;

[0049] Operational Status Monitoring Unit: Through a multi-dimensional signal acquisition module, it monitors the core operating parameters of each module in real time, establishes a threshold library of normal operating parameters for each module, and quickly identifies abnormal states through parameter comparison;

[0050] Fault handling and alarm unit: Built-in fault level classification logic, triggers corresponding processing mechanisms for different levels, and provides audible and visual alarms.

[0051] Compared with the prior art, the advantages of the present invention are:

[0052] Employing a contactless image acquisition mode, this system utilizes wide-angle camera coverage and dynamic tracking technology to avoid the invasiveness and cross-infection risks of traditional contact screening, making it suitable for all age groups. Leveraging the MediaPipe Holistic deep learning model, it accurately extracts key points on the human body, combining time-series data processing and multi-dimensional quantitative feature analysis to achieve objective and accurate assessment of motor coordination, eliminating the biases of traditional subjective judgments. Through personal historical data management and trend analysis, it can identify the risk of motor function decline in advance, enabling early screening and intervention. Supporting multi-scenario and age-group parameter customization, it boasts strong compatibility. The system ensures stable operation through time-series scheduling and data verification, and data security is guaranteed by encrypted storage mechanisms, providing an efficient and reliable motor coordination assessment solution for community health screening, medical institution diagnosis and treatment, and other scenarios. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the system proposed in this invention;

[0054] Figure 2This is a diagram of the contactless image acquisition module proposed in this invention;

[0055] Figure 3 This is a diagram of the human body key point extraction module proposed in this invention;

[0056] Figure 4 This is a diagram of the time series data processing module proposed in this invention;

[0057] Figure 5 This is a diagram of the quantization feature calculation module proposed in this invention;

[0058] Figure 6 This is a diagram of the personal historical data management module proposed in this invention;

[0059] Figure 7 This is a functional degradation trend module diagram proposed in this invention;

[0060] Figure 8 This is a diagram of the screening parameter configuration module proposed in this invention;

[0061] Figure 9 This is a diagram of the system control module proposed in this invention. Detailed Implementation

[0062] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0063] See Figure 1 As shown, a computer vision-based non-contact screening system for motion coordination includes:

[0064] Non-contact image acquisition module: Captures a continuous sequence of motion images of the person being screened using a non-contact camera, and performs noise reduction and distortion correction on the acquired images;

[0065] Human body key point extraction module: The acquired motion image sequence is analyzed frame by frame using a deep learning-based computer vision algorithm to extract the coordinates of the human body key nodes of the screened person in each frame, forming a set of human body key points corresponding to each frame.

[0066] Time series data processing module: It associates and integrates the set of human key points in chronological order to generate a human key point time series dataset. After smoothing and de-jittering, it fills in the missing data through time series interpolation to form standardized human key point time series data.

[0067] Quantitative Feature Calculation Module: Based on standardized human key point time series data, it calculates multi-dimensional quantitative feature indicators that characterize motion coordination, including trajectory features, rhythm and speed features, stability features, coordination features, and spatiotemporal features.

[0068] Personal historical data management module: Stores quantitative characteristic index data of the screened individuals, establishes unique personal exercise history data for each screened individual, and classifies and manages data based on the screened individual's identity identifier;

[0069] Functional decline trend module: The quantitative characteristic indicators of this screening are compared with the individual's exercise history data over time to obtain the changing trend of each quantitative characteristic indicator and determine whether there is a trend of motor function decline.

[0070] Screening parameter configuration module: used to preset and dynamically adjust the core parameters in the screening process, and to customize parameter combinations according to different age groups and different screening scenarios;

[0071] System control module: Establishes communication connections with each module, has built-in module scheduling algorithms and timing control logic, coordinates the working sequence of each module, controls the start, pause, termination and restart of the screening process, and realizes real-time data transmission and interactive verification between modules.

[0072] See Figure 2 As shown, the contactless image acquisition module specifically includes:

[0073] Camera unit: Employs a depth camera, positioned directly in front of the screening area, to capture a continuous sequence of motion images of the person being screened in real time;

[0074] Image preprocessing unit: Built-in dedicated image processing chip, which uses adaptive noise reduction algorithm to specifically eliminate environmental noise and electronic noise, and performs geometric distortion correction based on preset lens calibration parameters;

[0075] Dynamic tracking unit: Automatically optimizes the detection accuracy of key points in occluded areas.

[0076] Specifically, the image preprocessing unit processes the raw image stream captured by the camera in real time. First, it removes environmental noise, such as light reflection spots and background clutter, through an adaptive median filtering algorithm. Then, it performs geometric distortion correction based on preset lens calibration parameters to correct edge image stretching distortion caused by the wide-angle lens. Simultaneously, it performs quality verification on the processed image, using grayscale histogram analysis to determine whether the image brightness meets the standard. If motion blur occurs, it automatically increases the camera frame rate to 30fps.

[0077] When the person being screened completes the designated action or reaches the preset collection time, the system control module issues a collection termination command; the contactless image acquisition module stops image capture, encapsulates the preprocessed continuous image sequence into a standardized data format according to the timestamp order, and completes a single collection process.

[0078] See Figure 3 As shown, the human body key point extraction module specifically includes:

[0079] Image sequence receiving unit: Connects to the contactless image acquisition module to receive the processed motion image sequence of the screened person in real time, and uses a first-in-first-out buffering mechanism for caching;

[0080] Deep learning parsing unit: Based on the deep learning model in Azure Kinect Body Tracking SDK, it parses the image sequence frame by frame and extracts the three-dimensional coordinates of 32 key points of the whole body, including face, posture and hands;

[0081] Key point verification and optimization unit: It determines the occlusion status by the correlation of key point positions in adjacent frames, removes invalid coordinates caused by occlusion, optimizes key point coordinates by using a temporal smoothing filtering algorithm, corrects random errors, and forms a set of human body key points corresponding to each frame.

[0082] Specifically, the pre-trained MediaPipe Holistic model is loaded and initialized; the cached image sequence is imported into the model frame by frame, and the model extracts features from each frame to locate the face, body pose, and hand regions. Using the model's built-in keypoint detection algorithm, 32 key nodes across the entire body are accurately identified, and the 3D spatial coordinates of each keypoint are calculated simultaneously, outputting the original set of keypoint coordinates for each frame.

[0083] The key point coordinate data of adjacent frames are retrieved, and the occlusion status is determined by comparing the positional differences of the same key point. If the positional deviation of the same key point in adjacent frames exceeds the reasonable range, it is determined to be occlusion, and the invalid coordinates of the corresponding key point in that frame are removed. The remaining valid coordinates are processed by temporal smoothing filtering to weaken the random fluctuation error of the coordinates between frames and correct abnormal offset data. Finally, the optimized data is integrated to form a continuous and stable set of human body key points that corresponds one-to-one with each frame image.

[0084] See Figure 4 As shown, the time series data processing module specifically includes:

[0085] Temporal correlation and integration unit: Based on the timestamp synchronization algorithm, the coordinates of key human body points frame by frame are correlated and bound according to the acquisition time sequence, generating a structured time series dataset containing the three-dimensional spatial coordinates and visibility parameters of each key point and timestamp;

[0086] Data optimization unit: Adaptive temporal de-jitter algorithm is used to eliminate random noise in key point coordinates. For missing data caused by occlusion, temporal interpolation is performed to supplement it based on effective data from previous and next frames.

[0087] Data output adaptation unit: Converts the optimized standardized time series dataset into a format compatible with the quantization feature calculation module and transmits it in real time.

[0088] Specifically, based on the verified frame-by-frame data, the time-series correlation engine is started to sort the key point sets of each frame in ascending order of timestamps and construct the key point-time mapping relationship; for the 32 full-body key points, time series containing three-dimensional coordinates (x, y, z) and visibility parameters (v) are established respectively. Each series contains the continuous timestamps and three-dimensional coordinate change records of the corresponding key point in the whole acquisition cycle.

[0089] A sliding window traversal algorithm is used to scan the time-series dataset for missing data, locating regions with missing key points in a single frame or a small number of missing points in multiple consecutive frames. For local missing points in a single frame, linear interpolation is used to supplement the missing values ​​based on the coordinates of key points with the same ID in adjacent valid frames. For a small number of missing points in 2-3 consecutive frames, a strategy of copying neighboring frames combined with trend prediction is used to supplement the data. If more than 3 consecutive frames are missing, they are marked as data break regions and their locations are recorded.

[0090] See Figure 5 As shown, the quantization feature calculation module specifically includes:

[0091] Trajectory feature unit: The smoothness of the motion path is obtained by calculating the variance of the curvature change of the path curve; the accuracy of the motion execution is measured by the deviation value between the actual motion path and the standard motion path; and the presence of tremor during the motion is judged by the instantaneous displacement fluctuation amplitude of key points.

[0092] Rhythm and speed feature unit: The movement cycle is calculated based on the time interval between two consecutive completions of the same type of movement, and the speed symmetry and acceleration symmetry of the corresponding movements of the left and right limbs are calculated by the ratio of the peak speed and acceleration of the corresponding key points of the bilateral limbs.

[0093] Stability feature unit: Fit the center of gravity position by the coordinates of key points of the hip and torso, calculate the displacement range of the center of gravity in three-dimensional space, and measure the posture stability by the change in the center of gravity offset and the body tilt angle during single-leg support.

[0094] Coordination feature unit: The synchronicity of the left and right limb movements is assessed by the time difference of the key points of the bilateral limbs reaching the extreme position of the movement, and the symmetry of the left and right limb movements is calculated by the similarity of the movement amplitude and movement trajectory of the corresponding limbs on both sides.

[0095] Spatiotemporal characteristic unit: including the total time to complete the preset specified action, the number of times the action is completed per unit time, the time proportion of each action stage, and the connection time of action transition.

[0096] Specifically, the trajectory feature calculation extracts the complete motion path of each key point, generates a trajectory curve by concatenating the coordinates of consecutive frames, analyzes the curve fluctuation pattern to determine the smoothness, compares the overlap between the actual trajectory and the preset standard motion trajectory to quantify the motion accuracy, and monitors the displacement fluctuation amplitude of key points within 3 consecutive frames in real time to determine whether there is tremor.

[0097] The formula for calculating trajectory smoothness is:

[0098]

[0099] in, This represents the trajectory smoothness; the smaller the value, the smoother the motion path. To effectively acquire frames, Let be the curvature of the keypoint trajectory curve in the i-th frame. This represents the average value of the trajectory curvature over the entire acquisition period for N frames.

[0100] Rhythm and speed feature calculation: Locate the repetitive cycle nodes of similar movements, such as the peak values ​​of two leg raises in alternating leg raises, and calculate the time interval and fluctuation range of adjacent cycles; Derive the motion speed and acceleration of each key point by the coordinate difference of consecutive frames, and analyze the fluctuation pattern of its time sequence curve; Compare the peak speed and acceleration of key points of corresponding limbs on both sides and the occurrence time to quantify symmetry;

[0101] Stability feature calculation is based on fitting the human body's center of gravity position to the coordinates of key points on the hip and torso, tracking the three-dimensional spatial displacement trajectory of the center of gravity throughout the entire acquisition cycle, and statistically analyzing the maximum sway amplitude; for single-leg standing movements, the focus is on analyzing the continuous changes in the center of gravity offset and body tilt angle during the support period to quantify posture stability.

[0102] Coordination feature calculation records the time difference between the key points of corresponding limbs on both sides reaching the extreme value of movement to assess the synchronicity of movement; compare the similarity of the movement amplitude and trajectory curve of the two sides to quantify the symmetry of movement; analyze the start time difference and movement amplitude matching degree of adjacent joints to determine the multi-joint coordination efficiency.

[0103] The spatiotemporal characteristics are calculated and statistically analyzed to determine the total time span from the start to the completion of an action; the number of times an action is completed is counted per unit of time; the entire process of an action is broken down into the start, execution, and closing stages, and the time percentage of each stage and the transition time between stages are calculated.

[0104] See Figure 6 As shown, the personal historical data management module specifically includes:

[0105] Tiered storage unit: Equipped with a hybrid storage architecture consisting of SSD arrays and mechanical hard drives. The SSD array stores the screening data that has been accessed frequently in the past 3 years, while the mechanical hard drives archive long-term historical data, and independent data partitions are created according to the identity of the screened person.

[0106] Encryption and Security Unit: Integrates an AES-256 encryption module to encrypt stored data throughout the process and implements access control based on role-based access permissions;

[0107] Data interaction and scheduling unit: Supports querying historical data by identity and time range, and can link with the function degradation trend module to push historical data for a specified time period.

[0108] Specifically, the archived personal data is encrypted using the AES-256 encryption algorithm, covering characteristic indicator values, basic personal information, and screening auxiliary tags; local storage uses an industrial-grade solid-state drive array to store nearly 6 months of frequently accessed data, while cloud storage uses an encrypted cloud server to back up all historical data.

[0109] It supports multi-condition combined queries, allowing users to accurately retrieve historical data by criteria such as identity, screening time range, action type, and feature indicator type. Query results are returned in a standardized format, including complete feature indicator details, corresponding screening scenarios, and data processing logs. When a screened individual completes a repeat screening, the module automatically triggers a data update process, appending the new screening data to the corresponding personal file according to archiving rules.

[0110] See Figure 7 As shown, the functional degradation trend module specifically includes:

[0111] Time-series comparative analysis unit: This unit performs time-series comparative analysis between the quantitative characteristic indicators obtained from this screening and the historical indicator data in the individual's historical data archive.

[0112] Trend identification unit: By calculating the rate of change, fluctuation range and trend fit of each quantitative characteristic indicator within a continuous screening period, a personal motor function change curve is constructed, and the presence of a motor function degeneration trend is identified based on the curve;

[0113] Early warning unit: When the change range of one or more quantitative characteristic indicators exceeds the preset normal fluctuation threshold, or the change trend shows a continuous decline, a functional degradation early warning signal is generated. The early warning signal includes the abnormal characteristic indicator type, degradation degree assessment and change trend.

[0114] Data output unit: Outputs trend analysis results and change curve data, and synchronously saves analysis logs to personal exercise history data.

[0115] Specifically, based on the identity of the screened individuals and the screening timestamp, a two-dimensional correlation model of time and feature indicators is constructed. The historical data after screening is sorted in chronological order of screening time to form a continuous historical feature time series. Using the current screening data as a benchmark, the historical screening cycles are divided according to the same time span to achieve time alignment between the current data and the data of each historical cycle. To address the possible differences in the collection frame rate between different screening cycles, the data density is unified through data interpolation.

[0116] The multi-threaded parallel comparison engine is launched to perform time-series change analysis on five categories of feature indicators: First, single indicator trend tracking, calculating the change range of each feature indicator in the continuous screening period to locate indicators with abnormal numerical fluctuations; second, cross-indicator correlation analysis, exploring the correlation changes of features in different dimensions; and third, individual baseline calibration, using the average of the feature indicators of the screened person's previous three consecutive normal screenings as a benchmark to establish an individual's exclusive normal fluctuation range.

[0117] Based on the comparative analysis results, a trend judgment rule base is constructed, and a functional degradation trend is identified through a triple judgment logic: when one or more characteristic indicators show a continuous decline for two or more consecutive screening cycles; or when the change of a single indicator exceeds 1.5 times the individual's normal fluctuation threshold; or when there is a synergistic degradation correlation across indicators, it is judged that there is a trend of motor function degradation; according to the number of degradation indicators, the magnitude of change and the duration, the warning level is divided into three levels: mild, moderate and severe.

[0118] See Figure 8 As shown, the screening parameter configuration module specifically includes:

[0119] Hierarchical parameter management unit: Based on the hierarchical storage of parameters, the data acquisition class includes resolution and frame rate, the feature calculation class includes smoothness threshold, and the early warning class includes degradation fluctuation threshold.

[0120] Scenario-based parameter adaptation unit: Built-in multi-scenario preset parameter templates, including community health screening, medical institution special screening, and elderly health monitoring. The templates optimize parameter combinations for different age groups.

[0121] Parameter synchronization and verification unit: pushes configuration parameters to each module in real time through a clock synchronization mechanism, and verifies the module's operating status after the parameters are applied.

[0122] Specifically, it supports users in selecting screening scenarios (community health screening / medical institution-specific screening / elderly health monitoring) and the attributes of the screened group (age group: children / youth / middle-aged / elderly; basic physical condition: healthy / chronic disease / postoperative rehabilitation). Based on a preset scenario-parameter mapping rule library, it generates adaptive parameter combinations. For example, community health screening is adapted to low-complexity actions and medium acquisition frame rate; medical institution-specific screening increases the frame rate to 30fps and adds complex action types such as bending over and standing up, and continuous squatting and standing up; elderly health monitoring simplifies the action process, extends the action completion time, and reduces the action intensity.

[0123] See Figure 9 As shown, the system control module specifically includes:

[0124] Timing scheduling unit: Based on the phased development of the screening process, a module coordination timing table is formulated. The working rhythm of each module is coordinated through clock synchronization signals. By adjusting the timing matching relationship between contactless image acquisition and key point extraction, a time-coherent data transmission link is obtained.

[0125] Data interaction verification unit: It establishes a bidirectional communication link with each module through a multi-channel data interface matrix, uses the CRC check algorithm to verify cross-module transmitted data, and triggers a retransmission mechanism for abnormal data;

[0126] Operational Status Monitoring Unit: Through a multi-dimensional signal acquisition module, it monitors the core operating parameters of each module in real time, establishes a threshold library of normal operating parameters for each module, and quickly identifies abnormal states through parameter comparison;

[0127] Fault handling and alarm unit: Built-in fault level classification logic, triggers corresponding processing mechanisms for different levels, and provides audible and visual alarms.

[0128] Specifically, the system receives a combination of adaptation parameters and generates a refined time-series schedule based on the entire "acquisition-processing-analysis-output" logic. After the process starts, a start command is first sent to the contactless image acquisition module, simultaneously triggering the light intensity monitoring of the environment adaptation unit. When the acquisition module sends back the "image sequence ready" signal, a data parsing command is sent to the human body key point extraction module. Subsequently, each module is triggered in sequence according to the order of "time series data processing → multi-dimensional quantitative feature calculation → personal historical data management → motor function degradation trend analysis," with the trigger interval between adjacent modules dynamically adjusted according to the acquisition frame rate. If a module sends back "data processing in progress," the triggering of the next level module is automatically paused until a "processing complete" signal is received.

[0129] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0130] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0131] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A computer vision based motion coordination touchless screening system, characterized in that, Comprise: Non-contact image acquisition module: capture continuous motion image sequence of the screened person through non-contact camera, and denoise and correct distortion of the collected image; Human body key point extraction module: analyze the collected motion image sequence frame by frame through computer vision algorithm based on deep learning, extract the human body key point coordinates of each frame of image, and form the corresponding human body key point set frame by frame; Time series data processing module: correlate and integrate the human body key point set in time sequence to generate human body key point time series data set, supplement missing data through time series interpolation after smoothing and de-bouncing, and form standardized human body key point time series data; Quantitative feature calculation module: based on the standardized human body key point time series data, calculate multi-dimensional quantitative feature indexes representing motion coordination, including trajectory feature, rhythm and speed feature, stability feature, coordination feature and space-time feature; Personal history data management module: store the quantitative feature index data of the screened person, establish a unique personal motion history data for each screened person, and manage the data based on the identity of the screened person; Functional degradation trend module: compare the quantitative feature indexes of this screening with the personal motion history data in time sequence, obtain the change trend of each quantitative feature index, and judge whether there is a motion function degradation trend; Screening parameter configuration module: used for presetting and dynamically adjusting the core parameters in the screening process, and customizing parameter combination according to different age groups and different screening scenes; System control module: communicate with each module, built-in module scheduling algorithm and time sequence control logic, overall plan the working time sequence of each module, control the start, pause, termination and restart of the screening process, realize real-time transmission and interaction verification of data among modules.

2. The computer vision based motor coordination touchless screening system of claim 1, wherein, The non-contact image acquisition module specifically comprises: Camera unit: a depth camera is used, which is arranged in front of the screening area to capture continuous motion image sequence of the screened person in real time; Image preprocessing unit: built-in special image processing chip, specifically eliminate environmental noise and electronic noise through adaptive noise reduction algorithm, and correct geometric distortion based on preset lens calibration parameters; Dynamic tracking unit: automatically optimize the key point detection accuracy of the occlusion area.

3. The computer vision based motor coordination touchless screening system of claim 1, wherein, The human body key point extraction module specifically comprises: Image sequence receiving unit: interface with the non-contact image acquisition module, receive the processed motion image sequence of the screened person in real time, and use first-in-first-out cache mechanism for caching; Deep learning analysis unit: based on the deep learning model in Azure Kinect Body Tracking SDK, analyze the image sequence frame by frame, and extract 32 three-dimensional coordinates of whole body key points including face, posture and hand; Key point verification and optimization unit: judge the occlusion state through the correlation of adjacent frame key point positions, eliminate invalid coordinates caused by occlusion, optimize key point coordinates through time sequence smoothing filter algorithm, correct random errors, and form the corresponding human body key point set frame by frame.

4. The computer vision based motor coordination touchless screening system of claim 1, wherein, The time series data processing module specifically comprises: The time sequence correlation integration unit: based on a timestamp synchronization algorithm, the frame-by-frame human body key point coordinates are correlated and bound according to the acquisition time sequence, to generate a structured time sequence data set containing three-dimensional space coordinates and visibility parameters of each key point and a timestamp; The data optimization unit: an adaptive time sequence de-bouncing algorithm is used to eliminate random noise of the key point coordinates, and for missing data caused by occlusion, time sequence interpolation is performed based on the effective data of the previous and next frames to supplement the missing data; The data output adaptation unit: the optimized standardized time sequence data set is converted into a format compatible with the quantified feature calculation module, and is transmitted in real time.

5. The computer vision based motor coordination touchless screening system of claim 1, wherein, The quantified feature calculation module specifically includes: The trajectory feature unit: the smoothness of the motion path is calculated by the variance of the curvature change of the path curve, the accuracy of the action execution is measured by the deviation value of the actual motion path and the standard action path, and whether there is tremor in the motion process is judged by the instantaneous displacement fluctuation amplitude of the key point; The rhythm and speed feature unit: the action period is calculated based on the time interval of the completion of the same type of action twice, and the speed symmetry and acceleration symmetry of the corresponding actions of the left and right limbs are calculated by the speed and acceleration peak value ratio of the corresponding key points of the left and right limbs; The stability feature unit: the barycenter position is fitted by the hip and trunk key point coordinates, the displacement range of the barycenter in the three-dimensional space is calculated, and the posture stability is measured by the barycenter offset and the change amplitude of the body inclination angle during single leg support; The coordination feature unit: the synchronization of the left and right limbs is evaluated by the time difference of the key points of the left and right limbs reaching the motion extreme position, and the symmetry of the left and right limbs is calculated by the motion amplitude and the motion trajectory similarity of the corresponding limbs; The space-time feature unit: including the total time of completing the preset specified action, the number of action completions per unit time, the time proportion of each action stage, and the connection time of action conversion.

6. The computer vision based motor coordination touchless screening system of claim 1, wherein, The personal history data management module specifically includes: The hierarchical storage unit: a hybrid storage architecture composed of an SSD array and a mechanical hard disk is carried, the SSD array stores the screening data of recent 3 years with high frequency access, the mechanical hard disk archives long-term historical data, and an independent data partition is established according to the identity of the screened person; The encryption and security unit: an AES-256 encryption module is integrated, the stored data is encrypted throughout, and access control is performed based on role-based access permissions; The data interaction and scheduling unit: supports querying historical data according to identity, time range, and data linkage with the function degradation trend module to push historical data of a specified time period.

7. The computer vision based motor coordination touchless screening system of claim 1, wherein, The function degradation trend module specifically includes: The time sequence comparison and analysis unit: the quantified feature indicators obtained in this screening are compared and analyzed with the historical indicator data in the personal history data archive in time sequence; The trend identification unit: by calculating the change rate, fluctuation range and trend fitting degree of each quantified feature indicator in the continuous screening period, a personal motion function change curve is constructed, and based on the curve, whether there is a motion function degradation trend is identified; Early warning unit: When the change amplitude of one or more quantitative characteristic indicators exceeds the preset normal fluctuation threshold, or the change trend shows a sustained decline, a functional degradation early warning signal is generated, which includes the type of abnormal characteristic indicators, the degree of degradation assessment, and the change trend; Data output unit: output the trend analysis results and change curve data, and simultaneously store the analysis log in the personal exercise history data.

8. The computer vision based motor coordination touchless screening system of claim 1, wherein, The screening parameter configuration module specifically includes: Hierarchical parameter management unit: based on the hierarchical storage of parameters of acquisition class, feature calculation class and early warning class, the acquisition class contains resolution and frame rate, the feature calculation class contains smoothing threshold, and the early warning class contains degradation fluctuation threshold; Scenarized parameter adaptation unit: built-in multi-scene preset parameter templates, including community health screening, medical institution special screening and old-age care monitoring, the templates optimize parameter combinations for different age groups; Parameter synchronization and verification unit: through the clock synchronization mechanism, the configuration parameters are pushed to each module in real time, and the module running state after parameter application is verified.

9. The computer vision based motor coordination touchless screening system of claim 1, wherein, The system control module specifically includes: Timing scheduling unit: based on the screening whole process, the module coordination timing table is formulated in stages, the clock synchronization signal is used to coordinate the work rhythm of each module, the timing matching relationship of non-contact image acquisition and key point extraction is adjusted, and the timing coherent data transmission link is obtained; Data interaction verification unit: through the multi-channel data interface matrix, a bidirectional communication link is established with each module, the CRC verification algorithm is used to verify the cross-module transmission data, and the retransmission mechanism is triggered for abnormal data; Running state monitoring unit: through the multi-dimensional signal acquisition module, the core running parameters of each module are monitored in real time, the module normal running parameter threshold library is established, and the abnormal state is quickly identified through parameter comparison; Fault handling and alarm unit: built-in fault level division logic, corresponding processing mechanism is triggered for different levels, and sound and light alarm is performed.

Images