Pure vision-based skipping motion compliance detection method, device, equipment and medium

Abstract

The application provides a pure-vision-based skipping motion compliance detection method, device, equipment and medium, the method comprising: collecting real-time video stream of a skipping test area through a monocular camera, and generating a region coordinate configuration file defining the effective activity range of each tester; performing multi-target detection on the video stream frame by frame, synchronously outputting human key skeleton point coordinates and rope detection box information, and binding the rope target in each frame to the corresponding tester; obtaining time sequence characteristic data; based on the time sequence characteristic data, the region coordinate configuration file and a preset skipping rule, performing effective counting and anti-cheating determination on each skipping motion of each tester; outputting the effective skipping count, the violation type and the detailed time sequence data aligned with the video frame timestamp of each tester; reducing the deployment cost and ensuring the skipping count accuracy.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and in particular to a method, apparatus, device, and medium for detecting compliance of rope skipping movements based on pure vision. Background Technology

[0002] As a core component of middle school entrance examination physical education tests, school physical fitness training, mass fitness activities, and amateur competitions, the validity of jump rope scores and the compliance of movements (accuracy of counting, standardization of movements, and anti-cheating measures) directly affect the fairness of tests and competitions. In scenarios such as large-scale middle school entrance examinations, multi-person parallel training, and amateur competitions, traditional judgment methods rely on manual counting, mechanical counting equipment, or general vision solutions, which suffer from prominent problems such as large counting errors, weak anti-cheating capabilities, and lack of movement evaluation. With the intelligent development of AI technology in the sports field, there is an urgent need for a pure vision detection solution that requires no additional hardware, supports multi-person parallel detection, is accurate in real time, and conforms to training and testing rules, in order to meet the practical needs of automated scoring, fair competition, and efficient training evaluation.

[0003] Existing technical solutions: 1. Manual counting and judgment: The number of jumps is counted by the judges to observe whether there are any violations such as jumping without a rope, swinging without a rope, or crossing the boundary, and the standard of the movements is judged based on experience.

[0004] 2. Physical sensing devices: The smart jump rope handle with built-in acceleration sensor and ground pressure sensor are used to determine whether a jump rope has been completed by triggering the sensor signal.

[0005] 3. General visual tracking solution: Based on general algorithms such as target detection and IOU matching, the system tracks the human body or rope, judges the rope jumping action through simple inter-frame differences, and outputs the counting results.

[0006] Insufficiency of existing technology: 1. Manual method: Subjective error is significant, especially in high-speed rope skipping scenarios where missed counts and over-counts are frequent, and the standards for judging the standard of movement vary from person to person; it is extremely inefficient, requiring an equal number of judges for multiple people to test at the same time, which is not suitable for large-scale testing scenarios such as the middle school entrance examination; it lacks data traceability, the judgment results are not objectively supported, and disputes are difficult to arbitrate.

[0007] 2. Physical sensing solution: High deployment cost, requiring each tester to be equipped with a dedicated smart jump rope or the venue to be modified to install pressure sensing equipment; difficult to maintain, the sensor is easily interfered with by factors such as sweat and vibration, and the accuracy decreases after long-term use; poor scene adaptability, disrupting the natural state of training and competition, and is not suitable for temporary venues or large-scale centralized testing.

[0008] 3. General vision solution: Insufficient adaptability to multi-person scenarios, lack of dedicated test area calibration, which can easily lead to confusion in personnel counting; lack of compliance judgment capability, only outputs counting results, which cannot effectively identify violations such as cordless cheating, interruption timeout, and boundary crossing, making it difficult to meet the fairness requirements of examinations and competitions. Summary of the Invention

[0009] The technical problem to be solved by the present invention is to provide a method, device, equipment and medium for detecting rope skipping movements based on pure vision, thereby reducing deployment costs and ensuring accurate rope skipping counts.

[0010] In a first aspect, the present invention provides a method for detecting compliance of jump rope movements based on pure vision, comprising the following steps: S1. Acquire real-time video stream of the jump rope test area through a monocular camera, receive calibration input for the test area corresponding to each tester in the video stream through a graphical user interface, and generate a regional coordinate configuration file that defines the effective activity range of each tester. S2. Perform multi-target detection on the video stream frame by frame, and simultaneously output the coordinates of key human skeleton points and rope detection box information; enlarge the detected human target box according to a first preset ratio, and simultaneously generate a second preset ratio enlarged box larger than the original box for the detected rope target box; based on the nearest distance principle and cross-union ratio overlap verification, bind the rope target in each frame to the corresponding tester. S3. Based on the skeletal point coordinate sequence and rope frame coordinate sequence bound to the same tester in consecutive frames, perform temporal analysis to extract rope skipping cycle features, motion quantification features and event-related temporal features to obtain temporal feature data. S4. Based on the time-series feature data, the regional coordinate configuration file, and the preset rope skipping rules, perform a count validity determination and anti-cheating determination for each rope skipping action of each tester. S5. Output the effective jump rope count, violation type, and detailed timing data aligned with the video frame timestamp for each tester, and overlay the test area box, human body tracking box and skeleton connection line, as well as dynamically updated count and status information on the real-time video stream.

[0011] Secondly, the present invention provides a purely vision-based rope skipping motion compliance detection device, comprising: The acquisition and calibration module acquires real-time video streams of the jump rope test area through a monocular camera, receives calibration inputs for the test area corresponding to each tester in the video stream through a graphical user interface, and generates a regional coordinate configuration file that defines the effective activity range of each tester. The multi-target detection and association module performs multi-target detection frame by frame on the video stream, and simultaneously outputs the coordinates of key human skeleton points and rope detection box information; it enlarges the detected human target box according to a first preset ratio, and simultaneously generates a second preset ratio enlarged box larger than the original box for the detected rope target box; based on the nearest distance principle and cross-union ratio overlap verification, it binds the rope target in each frame to the corresponding tester. The temporal modeling module performs temporal analysis based on the skeletal point coordinate sequence and rope frame coordinate sequence bound to the same test subject in consecutive frames, extracting rope skipping cycle features, motion quantification features, and event-related temporal features to obtain temporal feature data. The multi-dimensional compliance judgment module, based on the time-series feature data, the regional coordinate configuration file, and the preset rope skipping rules, performs a count validity judgment and anti-cheating judgment on each rope skipping action of each tester. The output and visualization module outputs the effective jump rope count, violation type, and detailed time-series data aligned with the video frame timestamp for each tester. It also overlays the test area box, human body tracking box and skeleton connection line, as well as dynamically updated count and status information onto the real-time video stream.

[0012] Thirdly, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in the first aspect.

[0013] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0014] One or more technical solutions provided by this invention have at least the following technical effects or advantages: 1. No hardware dependency and extremely low deployment cost: Only a single camera and a regular computing terminal are required. No additional hardware such as smart jump ropes and pressure sensors are needed. No site modification is required. It can be quickly adapted to various venues such as large-scale physical education tests for the high school entrance examination, school training, and community fitness events (applicable to both temporary and fixed venues). 2. Comprehensive and accurate compliance judgment: It covers the judgment of "count validity and multiple types of cheating behavior". It can not only accurately filter invalid counts (such as swinging the rope without jumping or interruption timeout), but also accurately identify violations such as no-rope cheating, cheating by switching people, and crossing the boundary. It completely avoids the subjective error of manual counting and meets the fairness requirements of standardized testing. 3. Excellent real-time performance and support for multiple parallel testing: Single frame processing time ≤25ms, supports real-time detection of more than 30FPS; at the same time, through multi-area calibration and target binding technology, it can stably support parallel detection of 8 or more people (single camera coverage), meeting the needs of large-scale multi-person testing scenarios such as middle school entrance examination and campus tests. 4. Highly scalable: By adjusting the counting cycle threshold and cheating judgment rules, it can be quickly adapted to different scenarios such as the high school entrance examination physical education (tripping deducts points / no points) and mass fitness (counting priority).

[0015] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0017] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the device in Embodiment 2 of the present invention. Detailed Implementation

[0018] The overall concept of the technical solution in this application is as follows: This invention addresses the core characteristics of rope skipping—"high speed, short cycle, multiple participants, and diverse cheating scenarios"—by proposing an integrated pure vision solution that combines "visual acquisition, multi-target detection, rope skipping-specific time-series modeling, and multi-dimensional compliance judgment."

[0019] 1. Based on the real-time video stream of the rope skipping scene captured by a monocular camera, the key points of multiple test areas (adapting to multiple people in parallel) are marked through human-computer interaction to clarify the effective test range; 2. A lightweight multi-target detection algorithm is adopted to simultaneously realize the real-time detection of the human body (key skeletal points) and ropes, filter background interference, and output the human skeletal coordinates (such as ankles, shoulders, and elbows) and rope detection box information for each frame; 3. Design a time-series modeling network for rope skipping. Based on the detection data sequence of continuous frames, extract core motion features such as rope skipping cycle, take-off height, arm angle, and landing trajectory to solve the problems of rope tracking breakage and missed detection under high speed and short cycle. 4. Integrate jump rope training / testing rules to construct a "three-dimensional compliance judgment model": count validity judgment (based on the collaborative verification of the rope's winding cycle and the human body's take-off action to exclude invalid swings); anti-cheating judgment (detecting violations such as no-rope swings, cross-area jump rope, and interruption timeouts). 5. Outputs valid counts, violation types, and detailed time-series data for each tester end-to-end, requiring no additional hardware and adaptable to various scenarios such as large-scale testing and daily training.

[0020] Specifically as follows: 1. Site marking 1.1 Multi-test area calibration Through the human-computer interaction interface of the computing terminal, an independent and valid test area is marked for each tester to meet the needs of multi-person parallel testing, and a scenario configuration file is generated. Four vertices are selected for each area (top left of A1, top right of B1, bottom left of C1, bottom right of D1) to form a rectangular area (such as test area 1, test area 2, test area 3, ...), which serves as the core basis for determining boundary crossing cheating.

[0021] 2. Real-time detection of multiple targets Real-time multi-target detection is the core of jump rope data acquisition, requiring the simultaneous acquisition of position and morphological information of two core targets: the human body (skeleton points) and the rope, to provide high-quality data input for temporal modeling. This solution adopts a fusion architecture of "YOLOv5 + lightweight pose estimation" to balance detection accuracy and real-time performance.

[0022] 2.1 Deployment of Detection Model and Analysis of Results The detection model adopts the YOLOv5+ lightweight pose estimation architecture. After being fine-tuned with a special dataset, it is deployed on a computing terminal. The model input is a 1920×1080 resolution jump rope image, and the output is a feature tensor containing the rope state, skeletal point coordinates, and confidence scores.

[0023] 2.2 Target Optimization and Correlation Human bounding box optimization: The original detection box is enlarged by 1.1 times to cover the range of limb swings during rope skipping, avoiding tracking breaks caused by arm movements. Rope bounding box optimization: The original detection box is retained for trajectory calculation, while a 1.3 times enlarged bounding box is generated for cross-frame tracking association. Person-rope binding: Based on the "nearest distance principle + GIoU check", the rope bounding box closest to the human bounding box in each frame is bound to the corresponding tester, filtering out isolated ropes without bound to the human (background interference), solving the target confusion problem in multi-person scenes.

[0024] 2.3 Redundant Target Filtering Spatial filtering filters out human and rope targets (such as onlookers and background clutter) outside the calibrated test area coordinates; confidence filtering sets dynamic confidence thresholds (human ≥ 0.6, rope ≥ 0.5) to filter out low-confidence false detection targets.

[0025] 3. Time-series modeling for rope skipping The core of rope skipping time series modeling is to transform the detection data of discrete frames into structured periodic features and event chains, solve the technical pain points of "high speed and short cycle (single cycle < 1 second) and discontinuous trajectory", and accurately extract core parameters such as rope skipping cycle and take-off height.

[0026] 3.1 Time-series data input layer: Multi-source data association and synchronization Human skeleton time sequence: coordinates of key bone points (ankle, shoulder, elbow, etc.) sorted by frame number, smoothed by Kalman filtering (filling in occasionally missed bone points).

[0027] Rope timing sequence: The sequence of center coordinates of the rope frame bound to the tester, synchronously recording the rope state in each frame.

[0028] 3.2 Rope Skipping Timing Feature Engineering: Period and Motion Feature Extraction Based on the core characteristic of rope skipping, which is "periodic repetition (jump-rope wrap-landing)," three types of temporal features are constructed.

[0029] Periodic temporal characteristics. Rope skipping cycle calculation: Based on the ankle coordinate temporal sequence, the complete cycle of "landing → take-off → landing" is identified, and the frame difference between two adjacent landing nodes is the length of a single cycle.

[0030] Motion quantification features. Take-off height: Based on the ankle coordinate time sequence, calculate the difference between the highest point of the ankle and the ankle coordinate at landing within a single cycle; Arm angle: Based on the shoulder, elbow, and wrist bone points, calculate the arm bending angle for each frame using the vector angle formula (calculate separately for left and right and take the average), and extract the difference between the "minimum angle at take-off" and the "maximum angle at landing" (reflecting the arm swing amplitude).

[0031] Event-related temporal characteristics. Rope skipping event chain: Each cycle is divided into three core events: "starting point (t1, ankle leaves) → highest point of rope wrapping around the body (t2, minimum y-coordinate of rope frame) → landing point (t3, ankle touches)", forming temporal anchor points; Rope-human coordination characteristics: The temporal changes in the relative position of the rope frame and the human frame are calculated. When the rope frame moves from below the human frame to above it, it is marked as "rope wrapping around the body is completed", which serves as the core basis for the validity of the count.

[0032] 4. Multi-dimensional compliance assessment By combining the jump rope test / competition rules, a compliance judgment system of "counting validity - anti-cheating" is constructed. All judgment logic is based on the site's fixed and time-series modeling data to ensure that the results are accurate and traceable.

[0033] 4.1 Determination of Count Validity To determine whether a jump rope session counts as a valid score, the following three conditions must be met: 1. Complete cycle: The complete node chain "t1-t2-t3" must be successfully identified in the temporal modeling, and the length of a single cycle must be within the range of 3 to 25 frames (corresponding to a cycle of 0.1 to 0.83 seconds at a frame rate of 30 FPS, which conforms to the normal jump rope rhythm); 2. Rope wrap verification: Before the landing node (t3), the rope must complete a complete wrap around the body (the rope frame moves from below the body to above, and the GIoU verification passes), excluding invalid actions such as "swinging the rope without jumping"; 3. Area compliance: Both the take-off and landing nodes must be located within the designated test area, excluding jump ropes that cross boundaries (such as the tester jumping out of the rectangular area). If the above conditions are met, the count is incremented by 1, and details such as the cycle and take-off height of the jump rope session are recorded simultaneously.

[0034] 4.2 Anti-cheating and anomaly detection To address common cheating / abnormal scenarios in jump rope, a multi-dimensional verification logic is constructed; specifically: 1. Detection of cordless cheating: If only human jumping / landing actions are detected within 3 consecutive cycles, and no bound rope target is detected (confidence level <0.5), it is marked as "cordless cheating" and the counting of the test subject is suspended; 2. Interruption timeout determination: If the frame difference between two adjacent landing nodes is greater than 25 frames, it is determined as "rope skipping interruption". The number of interruptions is recorded. Actions during the interruption are not counted in the valid count. 3. Detection of cheating by switching people: The similarity is calculated based on the human skeleton features (shoulder width, hip height ratio). If the similarity between the tester's skeleton features and the initial features in a certain frame is <0.7 (and continues for more than 3 frames), it is marked as "cheating by switching people" and the current counting result is locked.

[0035] 5. Output and Visualization The system outputs core data for each tester end-to-end, including: valid counts, number of interruptions, cheating type; time-series data (skeletal points and rope positions per frame); and supports real-time visualization: overlaying test area boxes, counting results, and violation prompts (such as "out of bounds!" "Cheating without a rope!") into the video stream, and recording annotated test videos for review.

[0036] Example 1 like Figure 1 As shown, this embodiment provides a vision-based method for detecting compliance in rope skipping movements, including the following steps: S1. Acquire real-time video stream of the jump rope test area through a monocular camera, receive calibration input for the test area corresponding to each tester in the video stream through a graphical user interface, and generate a regional coordinate configuration file that defines the effective activity range of each tester. S2. Perform multi-target detection on the video stream frame by frame, and simultaneously output the coordinates of key human skeleton points and rope detection box information; enlarge the detected human target box according to a first preset ratio, and simultaneously generate a second preset ratio enlarged box larger than the original box for the detected rope target box; based on the nearest distance principle and cross-union ratio overlap verification, bind the rope target in each frame to the corresponding tester. S3. Based on the skeletal point coordinate sequence and rope frame coordinate sequence bound to the same tester in consecutive frames, perform temporal analysis to extract rope skipping cycle features, motion quantification features and event-related temporal features to obtain temporal feature data. S4. Based on the time-series feature data, the regional coordinate configuration file, and the preset rope skipping rules, perform a count validity determination and anti-cheating determination for each rope skipping action of each tester. S5. Output the effective jump rope count, violation type, and detailed timing data aligned with the video frame timestamp for each tester, and overlay the test area box, human body tracking box and skeleton connection line, as well as dynamically updated count and status information on the real-time video stream.

[0037] In this embodiment, preferably, the multi-target detection in step S2 adopts an architecture that integrates the YOLOv5 target detection network and the lightweight human pose estimation network, and outputs human skeleton points and rope target boxes simultaneously through a forward inference process.

[0038] In this embodiment, preferably, step S2 further includes: filtering out human and rope targets located outside the calibration test area based on the area coordinate configuration file; and filtering out targets with detection confidence lower than the threshold based on the dynamic confidence threshold, wherein the dynamic confidence threshold is determined based on the statistical distribution of detection confidence of similar targets in recent historical frames; if there is no detection confidence of similar targets in recent historical frames, a default value is set.

[0039] In this embodiment, preferably, in step S3: The jump rope cycle feature is the cycle duration identified based on the vertical coordinate sequence of key ankle points and defined by two adjacent landing event points. The motion quantification features include at least: the take-off height calculated based on the difference between the highest and lowest vertical coordinates of the ankle key points within a single cycle, and the arm swing amplitude calculated based on the shoulder, elbow, and wrist key points; The event-related timing features include at least: a rope skipping event chain that identifies the starting point, the highest point of the rope wrapping around the body, and the landing point, and a sequence describing the completion state of the rope detection frame moving from the lower region to the upper region relative to the human body detection frame within the same period.

[0040] In this embodiment, preferably, before constructing the time-series data model in step S3, the skeletal point coordinate sequence is smoothed by Kalman filtering to fill in the detection gaps.

[0041] In this embodiment, preferably, the validity determination of the count in step S4 must simultaneously meet the following conditions: (1) Cycle integrity: A complete rope skipping event chain containing the starting point, the highest point of the rope wrapping around the body and the landing point is successfully identified in the time series model, and its cycle duration is within the preset reasonable frame range; (2) Rope wrap verification: Before the landing event point, the event-related timing characteristics indicate that a rope wrap has been completed; (3) Regional compliance: The starting point and landing point are both located within the test area marked by the tester.

[0042] In this embodiment, preferably, the anti-cheating determination in step S4 includes at least one of the following types: (1) Determination of non-rope cheating: If the event-related timing features do not indicate that the rope has been wrapped around the body within N consecutive rope skipping cycles, it is determined to be non-rope cheating. (2) Interruption determination: If the frame difference between two adjacent landing event points exceeds a preset threshold, the rope skipping is determined to be interrupted; (3) Detection of cheating by switching people: During the test, the similarity between the human skeleton features of the current frame and the baseline skeleton features established for the tester is calculated periodically. If the similarity is lower than the preset threshold for more than M frames, it is determined to be cheating by switching people.

[0043] Based on the same inventive concept, this application also provides an apparatus corresponding to the method in Embodiment 1, as detailed in Embodiment 2.

[0044] Example 2 like Figure 2 As shown, this embodiment provides a vision-based rope skipping motion compliance detection device, including: The acquisition and calibration module acquires real-time video streams of the jump rope test area through a monocular camera, receives calibration inputs for the test area corresponding to each tester in the video stream through a graphical user interface, and generates a regional coordinate configuration file that defines the effective activity range of each tester. The multi-target detection and association module performs multi-target detection frame by frame on the video stream, and simultaneously outputs the coordinates of key human skeleton points and rope detection box information; it enlarges the detected human target box according to a first preset ratio, and simultaneously generates a second preset ratio enlarged box larger than the original box for the detected rope target box; based on the nearest distance principle and cross-union ratio overlap verification, it binds the rope target in each frame to the corresponding tester. The temporal modeling module performs temporal analysis based on the skeletal point coordinate sequence and rope frame coordinate sequence bound to the same test subject in consecutive frames, extracting rope skipping cycle features, motion quantification features, and event-related temporal features to obtain temporal feature data. The multi-dimensional compliance judgment module, based on the time-series feature data, the regional coordinate configuration file, and the preset rope skipping rules, performs a count validity judgment and anti-cheating judgment on each rope skipping action of each tester. The output and visualization module outputs the effective jump rope count, violation type, and detailed time-series data aligned with the video frame timestamp for each tester. It also overlays the test area box, human body tracking box and skeleton connection line, as well as dynamically updated count and status information onto the real-time video stream.

[0045] In this embodiment, preferably, the multi-object detection in the multi-object detection and association module adopts an architecture that integrates the YOLOv5 object detection network and the lightweight human pose estimation network, and outputs human skeleton points and rope target boxes simultaneously through a forward inference process.

[0046] In this embodiment, preferably, the multi-target detection and association module further includes: filtering out human and rope targets located outside the calibration test area based on the region coordinate configuration file; and filtering out targets with a detection confidence level lower than the threshold based on a dynamic confidence threshold, wherein the dynamic confidence threshold is determined based on the statistical distribution of detection confidence levels of similar targets in recent historical frames; if there is no detection confidence level of similar targets in recent historical frames, a default value is set.

[0047] In this embodiment, preferably, in the time series modeling module: The jump rope cycle feature is the cycle duration identified based on the vertical coordinate sequence of key ankle points and defined by two adjacent landing event points. The motion quantification features include at least: the take-off height calculated based on the difference between the highest and lowest vertical coordinates of the ankle key points within a single cycle, and the arm swing amplitude calculated based on the shoulder, elbow, and wrist key points; The event-related timing features include at least: a rope skipping event chain that identifies the starting point, the highest point of the rope wrapping around the body, and the landing point, and a sequence describing the completion state of the rope detection frame moving from the lower region to the upper region relative to the human body detection frame within the same period.

[0048] In this embodiment, preferably, before constructing the time-series data model in the time-series modeling module, the skeletal point coordinate sequence is smoothed by Kalman filtering to fill in the detection gaps.

[0049] In this embodiment, preferably, the count validity determination in the multi-dimensional compliance judgment module must simultaneously meet the following conditions: (1) Cycle integrity: A complete rope skipping event chain containing the starting point, the highest point of the rope wrapping around the body and the landing point is successfully identified in the time series model, and its cycle duration is within the preset reasonable frame range; (2) Rope wrap verification: Before the landing event point, the event-related timing characteristics indicate that a rope wrap has been completed; (3) Regional compliance: The starting point and landing point are both located within the test area marked by the tester.

[0050] In this embodiment, preferably, the anti-cheating determination in the multi-dimensional compliance judgment module includes at least one of the following types: (1) Determination of non-rope cheating: If the event-related timing features do not indicate that the rope has been wrapped around the body within N consecutive rope skipping cycles, it is determined to be non-rope cheating. (2) Interruption determination: If the frame difference between two adjacent landing event points exceeds a preset threshold, the rope skipping is determined to be interrupted; (3) Detection of cheating by switching people: During the test, the similarity between the human skeleton features of the current frame and the baseline skeleton features established for the tester is calculated periodically. If the similarity is lower than the preset threshold for more than M frames, it is determined to be cheating by switching people.

[0051] Since the apparatus described in Embodiment 2 of the present invention is an apparatus used to implement the method of Embodiment 1 of the present invention, those skilled in the art can understand the specific structure and variations of the apparatus based on the method described in Embodiment 1 of the present invention, and therefore will not be described again here. All apparatuses used in the method of Embodiment 1 of the present invention fall within the scope of protection of the present invention.

[0052] Based on the same inventive concept, this application provides an electronic device embodiment corresponding to Embodiment 1, as detailed in Embodiment 3.

[0053] Example 3 This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it can implement any of the implementation methods in Embodiment 1.

[0054] Since the electronic device described in this embodiment is the device used to implement the method in Embodiment 1 of this application, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the method described in Embodiment 1 of this application. Therefore, how the electronic device implements the method in the embodiment of this application will not be described in detail here. Any device used by those skilled in the art to implement the method in the embodiment of this application falls within the scope of protection of this application.

[0055] Based on the same inventive concept, this application provides a storage medium corresponding to Embodiment 1, as detailed in Embodiment 4.

[0056] Example 4 This embodiment provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it can implement any of the implementation methods in Embodiment 1.

[0057] Example 5 This embodiment provides a complete system implementation process, which is deployed on a computing terminal equipped with an NVIDIA GTX 3060 GPU and connected to a 1080P, 30FPS webcam.

[0058] S101: Site Marking and Initialization After the system starts, the operator uses a graphical user interface (GUI) to mark an independent rectangular test area for each test subject standing side-by-side in the live video feed captured by the camera. Specifically, the operator clicks on the top-left, top-right, bottom-left, and bottom-right vertices of each area in sequence. The system records the pixel coordinates of each vertex (e.g., test subject 1's area: [A1(100,200), B1(300,200), C1(100,600), D1(300,600)]) and saves the set of coordinates for all areas as a configuration file named config.json. This file defines the spatial basis for determining "area compliance" in subsequent steps.

[0059] S102: Real-time Detection and Correlation of Multiple Targets Model Deployment: The system loads a pre-trained fusion detection model. This model uses YOLOv5s as the backbone network, with a lightweight pose estimation subnetwork (HRNet, a simplified version of MobileNet-v2, used in this embodiment) connected in parallel at its output. The model is fine-tuned and trained using jump rope scene images containing 10,000 annotated human skeleton points (17 points, COCO format) and rope bounding boxes.

[0060] Detection and Analysis: For each frame of a 1920×1080 input image, the model output is: Human target: bounding box (x1, y1, x2, y2), confidence score conf_p, coordinates of 17 key points (kp_x, kp_y) and key point confidence scores.

[0061] Rope target: bounding box (x1, y1, x2, y2), confidence level conf_r.

[0062] Target optimization: For each body frame, its width and height are increased to 1.1 times the original size to ensure that the arms waving when jumping rope are included.

[0063] For each rope body bounding box, in addition to retaining the original bounding box R_orig for calculating the center point (cx, cy), a bounding box R_large, which is 1.3 times larger, is generated for cross-frame tracking association.

[0064] Target association and filtering: Spatial filtering: Traverse all detected human and rope frames, and filter out any frames whose center point is not within any calibrated test area polygon.

[0065] Confidence filtering: Calculate the 25th percentile P25_p of the confidence scores for all human detections and the 20th percentile P20_r of the confidence scores for rope detections within the last 30 frames. Use max(0.5, P25_p) as the confidence threshold for human detections in the current frame and max(0.3, P20_r) as the confidence threshold for rope detections, filtering out targets below the threshold.

[0066] Person-Rope Binding: For each filtered person bounding box H_i, calculate its Euclidean distance to the center point of all filtered rope original bounding boxes R_orig_j. Select the nearest rope R_orig_n and calculate the generalized intersection-union ratio (GIoU) ​​between H_i and R_large_n. If GIoU > 0.1, bind the rope R_n to the person H_i. Ropes not bound to anyone are considered background noise and are filtered out.

[0067] S103: Time Series Modeling for Rope Skipping The system maintains a time-series data buffer for each tracked test subject, storing the data from the most recent 2 seconds (60 frames).

[0068] Data preprocessing: A Kalman filter is applied to smooth the coordinate sequence of each human skeleton point (especially the ankle points ankle_l and ankle_r) to fill in occasional missed detections caused by occlusion.

[0069] Feature extraction: Periodicity detection: Analyze the time-series curve of the smoothed average vertical coordinates y_ankle of both ankles. Find local maxima (the lowest point of the ankle, i.e., the landing event). Define the frame sequence between two adjacent landing events as a jump rope period T_cycle. Calculate the frame length L_cycle of T_cycle.

[0070] Motion quantification: Jump height H_jump: Within one cycle T_cycle, calculate the difference between the minimum and maximum values ​​of y_ankle, and convert it to centimeters based on the calibration information.

[0071] Arm swing amplitude A_arm: Calculate the angle between the left and right arms (formed by the shoulder, elbow, and wrist points) in each frame, and take the average of the two to obtain θ_arm. Within one cycle, calculate the difference between the maximum and minimum values ​​of θ_arm.

[0072] Event Related: Within each cycle T_cycle, define: t1 (starting point): The inflection point where the y_ankle curve begins to decline significantly from a local maximum (landing).

[0073] t2 (highest point of rope wrapping): the minimum point of the y_rope coordinate of the center of the rope frame during this period.

[0074] t3 (landing point): The local maximum point of y_ankle (i.e., the next landing event).

[0075] Body wrapping status determination: During the period from t1 to t3, determine whether the center of the rope frame (cx_rope, cy_rope) has moved from below the lower edge (y2) of the human body frame H_i to above the upper edge (y1). If so, mark that the cycle has completed one "body wrapping".

[0076] S104: Multi-dimensional Compliance Assessment Validity determination of the count: For each identified cycle T_cycle, the following conditions must be met simultaneously to be counted as a valid jump rope session: 3 frames ≤ L_cycle ≤ 25 frames (adapting to a normal pacing of 0.1 seconds to 0.83 seconds).

[0077] The event correlation characteristics indicate that the "circumference" was completed within this period.

[0078] When events t1 and t3 occur, the center point of the corresponding human body frame is located within the rectangular area marked by the test subject.

[0079] Anti-cheating and anomaly detection: Cordless cheating: If a tester fails to meet condition 2 (no body wrapping) for three consecutive identified cycles, it is judged as "cordless cheating", the system will suspend its counting and issue an alarm.

[0080] Interruption determination: If the frame difference L_pause between two adjacent landing events t3 is greater than 75 frames (corresponding to 2.5 seconds), it is determined as "rope skipping interruption". The number of interruptions is recorded. Actions during the interruption period are not counted.

[0081] Player substitution cheating detection: Upon initialization for each tester ID, calculate the average skeletal feature vector V_base for the first 10 frames, including standardized ratios of (shoulder width / height) and (hip width / height). During subsequent tracking, calculate the cosine similarity S between the current frame's feature V_current and V_base every 15 frames. If S < 0.7 for more than 5 consecutive frames, it is considered "player substitution cheating," and the current count result for that ID is locked.

[0082] S105: Output and Visualization The system outputs data in real time via a JSON interface and displays it overlaid on the video feed. Graphic overlay: Draw a semi-transparent green rectangular area bounding box based on config.json; draw the magnified human body bounding box (blue) and skeletal lines for each tracked target; draw the original detection box (red) at the location of the bound rope.

[0083] Text information: Above each test area, the text in the format "ID: 01 | Count: 45 | Status: Normal" is dynamically displayed. When a violation occurs, the status changes to "Out of bounds!" or "Cheating without cord!" and is highlighted with a bright red flashing indicator.

[0084] Data recording: The system records the detection results of each frame, the event and feature data of each cycle, and each judgment result in a log file in a timestamp-aligned manner, and generates a result video containing all overlaid graphics for subsequent review and analysis.

[0085] In this embodiment, preferably, the dynamic confidence threshold is calculated using a sliding window exponentially weighted moving average method. For human targets, a confidence queue C_p of length W=50 frames is maintained. The threshold Th_p for the current frame is calculated using the following formula: Th_p = α * Th_p_previous + (1 - α) * percentile(C_p, 30), where α = 0.8 is the decay factor, and percentile(C_p, 30) represents the 30th percentile in the queue. An absolute lower limit Th_min_p = 0.4 is also set. The threshold Th_r for rope targets is calculated in a similar manner, with parameters percentile(C_r, 20) and Th_min_r = 0.25. This method smooths out short-term fluctuations in confidence levels.

[0086] In this embodiment, preferably, the skeletal feature vector V used for detecting cheating by switching characters includes not only the shoulder-to-width ratio and hip-to-width ratio, but also the ratio of arm length to torso length, and the relative positional relationship of feature points formed by the coordinates of typical joint points when standing still. Similarity calculation uses Mahalanobis distance to better adapt to the covariance variations of different human postures. The judgment threshold is adaptively set based on the covariance matrix of the feature distribution of the initial few frames.

[0087] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0088] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0089] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0090] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0091] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for compliance detection of jump rope movements based on pure vision, characterized in that, Includes the following steps: S1. Acquire real-time video stream of the jump rope test area through a monocular camera, receive calibration input for the test area corresponding to each tester in the video stream through a graphical user interface, and generate a regional coordinate configuration file that defines the effective activity range of each tester. S2. Perform multi-target detection on the video stream frame by frame, and simultaneously output the coordinates of key human skeleton points and rope detection box information; enlarge the detected human target box according to a first preset ratio, and simultaneously generate a second preset ratio enlarged box larger than the original box for the detected rope target box; based on the nearest distance principle and cross-union ratio overlap verification, bind the rope target in each frame to the corresponding tester. S3. Based on the skeletal point coordinate sequence and rope frame coordinate sequence bound to the same tester in consecutive frames, perform temporal analysis to extract rope skipping cycle features, motion quantification features and event-related temporal features to obtain temporal feature data. S4. Based on the time-series feature data, the regional coordinate configuration file, and the preset rope skipping rules, perform a count validity determination and anti-cheating determination for each rope skipping action of each tester. S5. Output the effective jump rope count, violation type, and detailed timing data aligned with the video frame timestamp for each tester, and overlay the test area box, human body tracking box and skeleton connection line, as well as dynamically updated count and status information on the real-time video stream.

2. The method according to claim 1, characterized in that, The multi-object detection in step S2 adopts an architecture that integrates the YOLOv5 object detection network and the lightweight human pose estimation network, and outputs human skeleton points and rope target boxes simultaneously through a forward inference process.

3. The method according to claim 1, characterized in that, Step S2 further includes: filtering out human and rope targets located outside the calibration test area based on the area coordinate configuration file; and filtering out targets with a detection confidence lower than the threshold based on the dynamic confidence threshold, wherein the dynamic confidence threshold is determined based on the statistical distribution of detection confidence of similar targets in recent historical frames; if there is no detection confidence of similar targets in recent historical frames, a default value is set.

4. The method according to claim 1, characterized in that, In step S3: The jump rope cycle feature is the cycle duration identified based on the vertical coordinate sequence of key ankle points and defined by two adjacent landing event points. The motion quantification features include at least: the take-off height calculated based on the difference between the highest and lowest vertical coordinates of the ankle key points within a single cycle, and the arm swing amplitude calculated based on the shoulder, elbow, and wrist key points; The event-related timing features include at least: a rope skipping event chain that identifies the starting point, the highest point of the rope wrapping around the body, and the landing point, and a sequence describing the completion state of the rope detection frame moving from the lower region to the upper region relative to the human body detection frame within the same period.

5. The method according to claim 1 or 4, characterized in that, Before constructing the time-series data model in step S3, the skeletal point coordinate sequence is smoothed using Kalman filtering to fill in detection gaps.

6. The method according to claim 1, characterized in that, The validity determination of the count in step S4 must simultaneously meet the following conditions: (1) Cycle integrity: A complete rope skipping event chain containing the starting point, the highest point of the rope wrapping around the body and the landing point is successfully identified in the time series model, and its cycle duration is within the preset reasonable frame range; (2) Rope wrap verification: Before the landing event point, the event-related timing characteristics indicate that a rope wrap has been completed; (3) Regional compliance: The starting point and landing point are both located within the test area marked by the tester.

7. The method according to claim 1, characterized in that, The anti-cheating determination in step S4 includes at least one of the following types: (1) Determination of non-rope cheating: If the event-related timing features do not indicate that the rope has been wrapped around the body within N consecutive rope skipping cycles, it is determined to be non-rope cheating. (2) Interruption determination: If the frame difference between two adjacent landing event points exceeds a preset threshold, the rope skipping is determined to be interrupted; (3) Detection of cheating by switching people: During the test, the similarity between the human skeleton features of the current frame and the baseline skeleton features established for the tester is calculated periodically. If the similarity is lower than the preset threshold for more than M frames, it is determined to be cheating by switching people.

8. A vision-based rope skipping motion compliance detection device, characterized in that, include: The acquisition and calibration module acquires real-time video streams of the jump rope test area through a monocular camera, receives calibration inputs for the test area corresponding to each tester in the video stream through a graphical user interface, and generates a regional coordinate configuration file that defines the effective activity range of each tester. The multi-target detection and association module performs multi-target detection frame by frame on the video stream, and simultaneously outputs the coordinates of key human skeleton points and rope detection box information; it enlarges the detected human target box according to a first preset ratio, and simultaneously generates a second preset ratio enlarged box larger than the original box for the detected rope target box; based on the nearest distance principle and cross-union ratio overlap verification, it binds the rope target in each frame to the corresponding tester. The temporal modeling module performs temporal analysis based on the skeletal point coordinate sequence and rope frame coordinate sequence bound to the same test subject in consecutive frames, extracting rope skipping cycle features, motion quantification features, and event-related temporal features to obtain temporal feature data. The multi-dimensional compliance judgment module, based on the time-series feature data, the regional coordinate configuration file, and the preset rope skipping rules, performs a count validity judgment and anti-cheating judgment on each rope skipping action of each tester. The output and visualization module outputs the effective jump rope count, violation type, and detailed time-series data aligned with the video frame timestamp for each tester. It also overlays the test area box, human body tracking box and skeleton connection line, as well as dynamically updated count and status information onto the real-time video stream.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images