Multi-view-angle-based multi-dimensional monitoring method and system for basketball pole-around motion

Abstract

The invention discloses a multi-view-angle-based multi-dimensional monitoring method and a multi-view-angle-based multi-dimensional monitoring system for basketball rod-around motion, and belongs to the technical field of computer vision and physical training monitoring. The objective of the invention is to solve the problems of low violation judgment accuracy and lack of quantitative motion indexes caused by a shielding blind area existing in single-view-angle monitoring. According to the technical scheme, the method comprises the following steps: synchronously acquiring video streams through at least two distributed cameras; human body key points, basketballs and marker posts are detected by adopting a lightweight deep learning model, and multi-view detection results are fused based on a majority voting mechanism; fourier stride frequency calculation and knee joint angle fluctuation analysis technologies are fused, and a multi-dimensional monitoring system capable of automatically recognizing various illegal behaviors such as rod winding missing, walking and wrist turning and synchronously quantifying stride frequency, joint angles and other indexes is constructed. The system is mainly used for automatic guidance and special physical examination of basketball pole winding training.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of computer vision and sports training monitoring, and particularly relates to a multi-dimensional monitoring method and system for basketball around-pole movement based on multi-view. BACKGROUND

[0002] Basketball around-pole training is a core training subject for improving the breaking, ball control stability and speed control ability of players. The evaluation of the training effect needs to consider both the standardization of the around-pole path and the quality quantization of the movement process. At present, the evaluation in this field mainly relies on the manual observation and judgment of coaches. This method has significant technical defects in practical application. First, manual observation is usually based on a single fixed view. In the process of rapid movement of players, the body, basketball and marker pole are easily blocked by each other, forming a visual blind area. Key violations such as missing the marker pole and over-shoulder are difficult to be fully captured. Existing data shows that the missed judgment rate of single-view observation can be as high as more than 30%, and the judgment accuracy is seriously insufficient. Second, the traditional evaluation method lacks objective and quantitative data output. Coaches can only provide qualitative conclusions of qualified or unqualified based on experience, and cannot accurately measure key biomechanical and kinematic indicators such as step frequency rhythm, knee joint stability and around-pole path deviation. This limits the pertinence and scientificity of training feedback, and makes it difficult for athletes to accurately locate and correct technical action shortcomings. Third, the manual evaluation mode relying on professional personnel is low in efficiency. The number of objects that a single coach can effectively monitor in a unit of time is extremely limited. In the application scenarios of large-scale campus sports examination and club batch training, there are practical difficulties such as high labor cost, difficulty in unifying evaluation standards and bottleneck of overall efficiency.

[0003] In the existing technology, some researches have tried to use computer vision-based automatic monitoring scheme to replace manual observation. However, these schemes still have limitations. Single-camera monitoring schemes are limited to two-dimensional plane information, and the inherent view blocking problem cannot be fundamentally solved, resulting in large spatial positioning errors. On the other hand, some multi-camera monitoring schemes mainly focus on the three-dimensional reconstruction of the movement trajectory of the player or the basketball, and fail to effectively analyze the deep movement characteristics such as movement rhythm and joint angle fluctuation, resulting in single monitoring dimension and difficulty in establishing the internal relationship between violation behavior and player action quality and body state, which cannot provide deeper attribution analysis for training improvement. Therefore, there is an urgent need in the current technical field for a comprehensive monitoring technology that can integrate multi-view spatial information and multi-dimensional movement characteristic analysis to realize automatic and accurate identification of violation behaviors in basketball around-pole training process, and simultaneously output quantifiable movement performance indicators, so as to meet the automation needs of professional and large-scale sports training and examination. SUMMARY

[0004] To solve the above technical problems, the application provides a multi-dimensional monitoring method and system for basketball pole-winding movement based on multiple perspectives, which realizes real-time monitoring performance of 25 FPS or above through a multi-perspective data fusion algorithm and lightweight model optimization, effectively solves technical bottlenecks such as blind areas and insufficient accuracy of rule violation determination in single-perspective pole-winding monitoring, and is suitable for professional scenarios such as basketball pole-winding training guidance and special physical ability assessment.

[0005] In one aspect, the application provides a multi-dimensional monitoring method for basketball pole-winding movement based on multiple perspectives, which includes:

[0006] At least two cameras are deployed to synchronously collect multi-path video streams of the pole-winding movement area;

[0007] The multi-path video streams are processed to detect human key points, basketballs and marker poles in each video, and the detection results of the multi-path video streams are fused to obtain fused target state information;

[0008] Step frequency analysis and joint angle analysis are performed based on the fused target state information, and whether there is a rule violation is determined based on the analysis results and the spatial relationship between targets;

[0009] Multi-dimensional monitoring information containing rule violation determination results and movement analysis results is output.

[0010] Optionally, deploying at least two cameras includes: distributing the cameras on both sides of the pole-winding movement path, and making the field of view coverage areas of the cameras overlap each other to eliminate monitoring blind areas; and using a timestamp alignment mechanism to synchronize the multi-path video streams.

[0011] Optionally, processing the multi-path video streams includes: using a cluster of lightweight deep learning models running in parallel to detect human skeletal key points, basketball bounding boxes and center points, and marker pole bounding boxes and segmentation masks from each video stream.

[0012] Optionally, fusing the detection results of the multi-path video streams includes: aligning the detection results of each video stream within the same time window; and using a majority voting strategy, when the detection results or rule violation determinations for the same target are consistent in more than a preset number of video streams, adopting them as the fused target state information.

[0013] Optionally, determining whether there is a rule violation includes: determining a missed-winding violation based on the distance between the human center point and the marker pole center point; determining a marker pole falling violation based on the inclination angle of the marker pole segmentation mask; and comprehensively determining a walking violation in combination with the step frequency analysis results and the joint angle analysis results.

[0014] Optionally, the step frequency analysis comprises: extracting a vertical displacement time sequence signal of a human ankle key point, and performing frequency domain transformation on the vertical displacement time sequence signal to extract a main frequency as a step frequency; and the joint angle analysis comprises: calculating a knee joint angle and counting a knee joint angle time sequence fluctuation degree.

[0015] Optionally, the outputting of the multi-dimension monitoring information comprises: rendering and displaying the fused target state information, a step frequency change curve, a joint angle fluctuation index and a violation warning mark in real time in a visual interface; and asynchronously storing a violation judgment result and an associated motion analysis result.

[0016] In another aspect to achieve the above object, the application further provides a multi-dimension monitoring system for basketball pole-winding motion based on multi-view, comprising:

[0017] a video acquisition synchronization module configured to deploy at least two cameras to synchronously acquire multi-channel video streams of a pole-winding motion area;

[0018] a target detection fusion module configured to process the multi-channel video streams to detect human key points, a basketball and a marker pole in each channel, and fuse detection results of the multi-channel video streams to obtain fused target state information;

[0019] a motion analysis and violation judgment module configured to perform step frequency analysis and joint angle analysis based on the fused target state information, and judge whether a violation behavior exists based on analysis results and a spatial relationship between targets;

[0020] a multi-dimension information output module configured to output multi-dimension monitoring information containing a violation judgment result and a motion analysis result.

[0021] An electronic device, comprising: a processor and a memory storing computer program instructions;

[0022] The processor executes the computer program instructions to implement the multi-dimension monitoring method for basketball pole-winding motion based on multi-view.

[0023] A computer storage medium storing computer program instructions, the computer program instructions being executed by a processor to implement the multi-dimension monitoring method for basketball pole-winding motion based on multi-view.

[0024] The technical effect of the present application: the present application discloses a kind of based on multi-view basketball around pole movement multidimensional monitoring method and system, by multi-view cooperation and multidimensional feature fusion, effectively overcome the visual obstruction problem in traditional single-view monitoring, significantly improve the comprehensive and accuracy of the determination of a variety of illegal behaviors such as missing flag pole, walking, wrist turning, etc..Meanwhile, the present application not only realizes the automatic identification of illegal behavior, but also provides quantitative training index reflecting the rhythm control and body stability of player by fusing step frequency analysis and joint angle fluctuation analysis, so as to deepen the evaluation from qualitative judgment to quantitative analysis level, which provides direct parameter basis for training improvement.In terms of practicality, the present application meets the real-time monitoring demand while ensuring high recognition accuracy by using lightweight model and parallel processing architecture, and the system can be deployed relying on general hardware, which reduces the threshold of large-scale application.In addition, the modular design enables the system to flexibly adapt to different number of cameras and adjustable determination threshold, so as to have good expansibility and be able to adapt to diversified training scenes and standard requirements. BRIEF DESCRIPTION OF DRAWINGS

[0025] The accompanying drawings, which form a part of this application, are intended to provide further understanding of the application and are incorporated herein in their entirety. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application. In the drawings:

[0026] Figure 1 A flowchart of a kind of based on multi-view basketball around pole movement multidimensional monitoring method of the embodiment of the present application is shown in the figure;

[0027] Figure 2 A structure diagram of a kind of based on multi-view basketball around pole movement multidimensional monitoring system of the embodiment of the present application is shown in the figure;

[0028] Figure 3 A multi-view camera deployment diagram of the embodiment of the present application is shown in the figure;

[0029] Figure 4 A Fourier step frequency calculation and illegal association analysis diagram of the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0030] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the drawings and in combination with embodiments.

[0031] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system, such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in different order from here.

[0032] As Figure 1As shown, the embodiment provides a multi-dimensional monitoring method for basketball pole-winding motion based on multi-view, which includes:

[0033] At least two cameras are deployed to synchronously collect multi-channel video streams of the pole-winding motion area;

[0034] The multi-channel video streams are processed to detect human key points, basketballs and marker poles in each video, and the detection results of the multi-channel video streams are fused to obtain fused target state information;

[0035] Step frequency analysis and joint angle analysis are performed based on the fused target state information, and whether there is a violation behavior is determined based on the analysis results and the spatial relationship between targets;

[0036] Multi-dimensional monitoring information containing violation determination results and motion analysis results is output.

[0037] Further, deploying at least two cameras includes: distributing the cameras on both sides of the pole-winding motion path, and making the field of view coverage areas of the cameras overlap each other to eliminate monitoring blind areas; and using a timestamp alignment mechanism to synchronize the multi-channel video streams.

[0038] Specifically, the implementation process of the embodiment includes:

[0039] As shown in Figure 3 2-4 USB 3.0 cameras (1080P resolution, 30 FPS frame rate) are deployed on both sides of the pole-winding path, synchronous collection is realized by a software timestamp alignment method (synchronization error ≤3 ms), the time consistency of multi-view data is ensured; each view frame image is scaled by an equal ratio according to the longest side ≤640px (maintaining the aspect ratio), image preprocessing is completed, and image noise interference is eliminated; data transmission delay is corrected by a timestamp alignment algorithm, accurate matching of multi-view frame data at the same time is ensured, and standardized synchronous image data is output.

[0040] Hardware selection: computing device: Intel i7-12700H central processor, RTX 3060 graphics processor (6GB video memory, supporting CUDA 11.7), 16GB DDR4 memory, 512GB solid state disk;

[0041] Camera: 4 USB 3.0 cameras (1080P resolution, 30 FPS frame rate, equipped with wide-angle lens);

[0042] Marker pole: 10 circular marker poles with a diameter of 10 cm and a height of 1.8 m, the surface is sprayed with a high-contrast yellow paint layer to improve recognition accuracy;

[0043] Auxiliary equipment: USB 3.0 hub (supporting 4-way synchronous data transmission), 5V trigger-type audible and visual alarm.

[0044] Site and equipment deployment: Around the pole site: Set up a rectangular training area with a length of 20m and a width of 7m, set up sign poles 5 meters in front of the starting line, a total of 5 rows, 2 poles in each row; The distance between the sign poles in each row is 3 meters, and the distance between the center points of the two sign poles in the same row is 1 meter, and it is 3 meters away from the same side line. The preset pole winding sequence is S-shaped path 1→2→3→4→5→6→7→8→9→10;

[0045] Camera deployment: Deploy 2 cameras on both sides of the pole winding path, 1.5m away from the path edge, installation height 1.2m, adjacent camera interval 5m, angle with path 60°, ensure that the monitoring area has no blind area;

[0046] Synchronous connection: 4 cameras are connected to the computing device through the USB 3.0 hub, and the audible and visual alarm is connected to the computing device through the GPIO interface, realizing the rapid response of the illegal information.

[0047] System calibration: Time synchronization calibration: Test the synchronization error of multiple cameras through software timestamp alignment, adjust the delay compensation parameter, ensure that the time difference of multi-view data is less than or equal to 3ms, and meet the accuracy requirement of 0.5s synchronization window;

[0048] Model parameter configuration: According to the actual scene, adjust the detection model confidence threshold, illegal judgment threshold (such as the leakage winding distance threshold 200 pixels, the walking step number threshold 1 step, the knee joint angle fluctuation threshold 10 degrees, etc.), Optimize the balance between detection accuracy and false detection rate.

[0049] Further, the processing of the multiple video streams includes: using a cluster of lightweight deep learning models running in parallel to detect human body key points, basketball bounding boxes and center points, and sign pole bounding boxes and segmentation masks from each video stream, respectively.

[0050] Specifically, the implementation process of the embodiment includes:

[0051] The cluster of lightweight deep learning models includes:

[0052] yolo11s-pose model: used to detect the coordinates of 17 key skeletal points of the human body, focusing on extracting motion feature points such as ankles, wrists, shoulders, and knees;

[0053] Basketball detection model: used to output the bounding box and center coordinates of the basketball, using custom anchor boxes to improve small target detection accuracy;

[0054] Signpost detection model: used to identify the signpost bounding box and segmentation mask, supports the detection of standard-shaped signposts such as circles and squares, and determines the inclination state of the signpost through mask analysis;

[0055] Multi-view data fusion unit: uses a time window synchronization mechanism (synchronization window 0.5s) to time-align the detection results of each view, and determines the final violation result through a majority voting mechanism (at least 2 views simultaneously detect a violation), effectively reducing the single-view false detection rate.

[0056] Cross-view target detection and data fusion: each view independently runs a lightweight model for parallel inference to improve detection efficiency, including: the yolo11s-pose model extracts 17 key skeletal point coordinates and confidence of the human body, focusing on extracting motion feature points such as ankles, wrists, shoulders, and knees; the basketball special detection model is based on custom anchor box design, outputting basketball bounding box, center coordinates and confidence, optimizing small target detection accuracy; the signpost detection model is trained through the YOLOv11 instance segmentation model architecture, accurately identifying the signpost bounding box and segmentation mask, supporting different shaped signposts such as circles and squares, and determining the inclination state of the signpost through mask analysis.

[0057] Further, the detection results of the fused multi-channel video stream include: aligning the detection results of each video stream within the same time window; using a majority voting strategy, when the detection results or violation judgments for the same target are consistent in more than a preset number of video streams, they are adopted as the fused target state information.

[0058] Specifically, the implementation process of the embodiment includes:

[0059] Multi-view data fusion optimization steps, specifically:

[0060] Time synchronization: uses a timestamp alignment mechanism to time-match multi-view detection results through a synchronization window (0.5s), ensuring that detection results at the same time can be correctly fused;

[0061] Majority voting mechanism: uses a majority voting strategy for each view detection result, and when at least 2 views simultaneously detect the same violation, it is determined as the final violation, effectively reducing the single-view false detection rate;

[0062] Real-time guarantee: uses a GPU parallel inference architecture (multiple models are calculated synchronously) and a "collection-detection-fusion-rendering" multi-thread asynchronous processing mechanism, through a frame skipping strategy (inference once every 5 frames) and a result caching mechanism, the total processing delay of a single frame is ≤40ms, meeting the 25FPS real-time monitoring demand, and improving the processing efficiency.

[0063] Further, the determination of whether there is a violation includes: determining the missing winding violation based on the distance between the human center point and the flagpole center point; determining the flagpole falling violation based on the inclination angle of the flagpole segmentation mask; and comprehensively determining the walking violation in combination with the step frequency analysis result and the joint angle analysis result.

[0064] Specifically, the implementation process of the embodiment includes:

[0065] In the violation determination step, the missing winding flagpole determination adopts a simplified method based on two-dimensional distance: the two-dimensional Euclidean distance between the human center point (human bounding box center) and the center point of each flagpole (flagpole bounding box center) is calculated in real time, and when the distance > 200 pixels, it is determined as a missing winding violation; the flagpole falling determination is realized by analyzing the inclination angle of the minimum circumscribed rectangle of the flagpole segmentation mask, and when the inclination angle > 45 degrees, it is determined as a flagpole falling violation.

[0066] The violation determination step specifically includes:

[0067] Missing winding flagpole determination: the two-dimensional distance between the human center point and the flagpole center point is calculated in real time, and when the distance > 200 pixels, it is determined as a missing winding violation;

[0068] Flagpole falling determination: the inclination angle of the minimum circumscribed rectangle of the flagpole segmentation mask is analyzed, and when the inclination angle > 45 degrees, it is determined as a flagpole falling violation;

[0069] Walking violation determination: in combination with displacement distance detection (step number > 1), Fourier step frequency analysis (0.5-3.0Hz effective frequency band) and knee joint angle fluctuation detection (standard deviation > 10 degrees), when the three are satisfied at the same time, it is determined as a walking violation;

[0070] Other violation determination: including wrist flipping (ball above wrist and arm vertical), double-hand ball handling (both hands holding the ball at the same time), below-knee ball touching (knee / ankle distance from the ball < 15 pixels), over-shoulder (holding the ball higher than the shoulder), etc.

[0071] Step frequency correlation violation analysis: when a violation occurs, if abnormal fluctuation (difference ≥ 20%) of step frequency or abnormal fluctuation of knee joint angle is simultaneously detected, it is marked as "active violation caused by motion deformation".

[0072] Violation judgment: Construct multi-condition correlation judgment logic, including: (1) Missing flag pole: Real-time calculation of the two-dimensional distance between the human body center point and the flag pole center point, when the distance > 200 pixels, it is judged as missing violation; (2) Flag pole down: By analyzing the minimum circumscribed rectangle angle of the flag pole segmentation mask, when the inclination angle > 45 degrees, it is judged as flag pole down violation; (3) Walking: Combined with displacement distance detection (step number > 1), Fourier step frequency analysis (0.5-3.0Hz effective frequency band) and knee joint angle fluctuation detection (standard deviation > 10 degrees), when the three are satisfied at the same time, it is judged as walking violation; (4) Other violations: Including wrist flipping (ball above wrist and arm vertical), double-hand dribbling (both hands holding the ball at the same time), below-knee touching the ball (knee / ankle and ball distance < 15 pixels), over-shoulder (holding the ball higher than the shoulder), etc.; When the violation occurs, if the step frequency abnormal fluctuation or the knee joint angle abnormal fluctuation is detected synchronously, it is marked as "active violation caused by action deformation".

[0073] Further, the step frequency analysis includes: extracting the vertical displacement time sequence signal of the human ankle key point, and performing frequency domain transformation on the vertical displacement time sequence signal to extract the main frequency as the step frequency; the joint angle analysis includes: calculating the knee joint angle and counting the knee joint angle time sequence fluctuation degree.

[0074] Specifically, the implementation process of the embodiment includes:

[0075] As shown in Figure 4 , in the Fourier step frequency calculation step, the y-axis component of the ankle two-dimensional coordinates of the last N frames (N = 3xFPS, about 3 second window) is intercepted to construct a time sequence signal, which is preprocessed by removing the mean value and then performs fast Fourier transform; Through 0.5-3.0Hz effective frequency band screening to remove noise interference, take the peak frequency in the frequency band as the real-time step frequency value; The judgment standard of step frequency abnormal fluctuation is that the absolute value of the difference between the current step frequency and the average step frequency in the last 3 seconds is ≥20%; At the same time, combined with the knee joint angle fluctuation analysis (standard deviation > 10 degrees in the last 2 second window) and displacement distance detection, walking violation is comprehensively judged.

[0076] Multi-dimensional motion analysis and violation detection: Fusion of two-dimensional detection data to realize motion parameter quantization and correlation analysis of violation behavior, specifically including: Fourier step frequency calculation: intercept the y-axis component of the ankle two-dimensional coordinates of the last N frames (N = 3xFPS, about 3 second window) to construct a time sequence signal, which is preprocessed by removing the mean value to eliminate baseline drift and then performs fast Fourier transform; Through 0.5-3.0Hz effective frequency band screening to remove noise interference, take the peak frequency in the frequency band as the real-time step frequency value; Calculate the difference between the current step frequency and the average step frequency in the last 3 seconds, when the absolute value of the difference is ≥20%, it is judged as step frequency abnormal fluctuation.

[0077] Knee joint angle fluctuation analysis: Real-time calculation of left and right knee joint angles (based on hip-knee-ankle three-point angle), maintenance of the angle history sequence in the last 3 seconds, calculation of the standard deviation in the last 2-second window, and determination of effective fluctuation when the standard deviation is greater than 10 degrees, which is used for comprehensive judgment of walking violations.

[0078] Further, the output multi-dimensional monitoring information includes: real-time rendering and display of the fused target state information, the step frequency change curve, the joint angle fluctuation index and the violation warning mark in the visualization interface; and asynchronous storage of the violation judgment result and the associated motion analysis result.

[0079] Specifically, the implementation process of the embodiment includes:

[0080] The two-dimensional visualization rendering unit realizes real-time display of multi-view pictures by using OpenCV technology, supports multi-view picture splicing display, real-time step frequency information display (left and right feet and average value), knee joint angle standard deviation display, violation type labeling and other functions; the asynchronous data storage unit stores violation images in the "violation_images / violation type / " directory according to the violation type, and saves violation logs as Excel files, which include core information such as violation time, violation type, whether holding the ball, dynamic threshold, ball radius, violation reason, and non-violation reason. The data storage delay is ≤3s and does not block the real-time monitoring process.

[0081] Integrating multi-view monitoring results, multi-view picture splicing display, real-time step frequency information display (left and right feet and average value), knee joint angle standard deviation display, violation type labeling and other visualization functions are realized by using OpenCV technology, and the visualization delay is ≤8ms; a "collection-detection-fusion-rendering" multi-thread asynchronous architecture is adopted to ensure that the total processing delay of a single frame is ≤40ms, meeting the real-time monitoring demand of 25FPS; data storage operations are executed by asynchronous threads, violation images are stored in the "violation_images / violation type / " directory according to the violation type, and violation logs are saved as Excel files, and the data storage delay is ≤3s and does not block the real-time monitoring process.

[0082] As shown in Figure 2 , the embodiment provides a multi-dimensional monitoring system for basketball pole-walking motion based on multi-view, which includes:

[0083] A video acquisition synchronization module is configured to deploy at least two cameras to synchronously acquire multi-channel video streams of the pole-walking motion area.

[0084] A target detection fusion module is configured to process the multi-channel video streams to detect human key points, a basketball and a marker pole in each video, and fuse the detection results of the multi-channel video streams to obtain fused target state information.

[0085] A motion analysis and violation judgment module is configured to perform step frequency analysis and joint angle analysis based on the fused target state information, and determine whether a violation exists based on the analysis results and the spatial relationship between targets.

[0086] A multi-dimensional information output module is configured to output multi-dimensional monitoring information including the violation judgment result and the motion analysis result.

[0087] Specifically, the multi-view video synchronization module in the embodiment includes a distributed camera unit, a time synchronization unit, and a frame standardization unit. The camera unit supports USB 3.0 / webcam access. The time synchronization unit achieves multi-device synchronization through hardware triggering or software alignment. The frame standardization unit completes image scaling, format conversion, and normalization processing to ensure the stability and consistency of input data.

[0088] The cross-view target detection module includes a multi-model inference unit, a multi-view data fusion unit, and a result caching unit. The multi-model inference unit runs human, basketball, and flagpole detection models in parallel. The multi-view data fusion unit achieves multi-view result fusion through time window synchronization and majority voting mechanism. The result caching unit improves processing efficiency through frame skipping strategy and result multiplexing mechanism, and provides high-quality detection data for subsequent analysis. The multi-model inference unit uses INT8 quantization processing lightweight model, and the flagpole detection model is trained through YOLOv11 instance segmentation model architecture.

[0089] The motion analysis and violation detection module integrates a Fourier step frequency calculation unit, a knee joint angle fluctuation analysis unit, multiple violation judgment units (walking, wrist flipping, double-hand dribbling, missing flagpole, flagpole falling, touching the ball below the knee, and over-the-shoulder), and a step frequency correlation analysis unit, to realize step frequency quantization, angle fluctuation evaluation, multiple violation identification, and behavior and parameter correlation analysis, and is the core logic layer of the system.

[0090] Multi-dimensional result output module: covering two-dimensional visualization rendering unit (OpenCV display), real-time state information display unit, asynchronous data storage unit. Through two-dimensional visualization technology, multi-view detection results and real-time motion parameters are displayed, the real-time state information display unit displays key indicators such as step frequency and knee joint angle standard deviation, and the asynchronous storage unit guarantees data retention without affecting the real-time performance of the system. The two-dimensional visualization rendering unit realizes the real-time display of multi-view pictures by using OpenCV technology, supports multi-view picture splicing display, real-time step frequency information display (left and right feet and average value), knee joint angle standard deviation display, violation type labeling and other functions; the asynchronous data storage unit stores the violation images according to the violation type in the 'violation_images / violation type / ' directory, and saves the violation log as an Excel file, which includes core information such as violation time, violation type, whether holding the ball, dynamic threshold, ball radius, violation reason, and non-violation reason, and the data storage delay is ≤3s and does not block the real-time monitoring process.

[0091] An electronic device, comprising: a processor and a memory storing computer program instructions;

[0092] The processor executes the computer program instructions to realize the multi-dimensional monitoring method for basketball around the pole movement based on multi-view.

[0093] A computer storage medium, the computer storage medium stores computer program instructions, the computer program instructions are executed by the processor to realize the multi-dimensional monitoring method for basketball around the pole movement based on multi-view.

[0094] The application discloses a multi-dimensional monitoring method and system for basketball around the pole movement based on multi-view, which effectively overcomes the visual occlusion problem in traditional single-view monitoring through multi-view cooperation and multi-dimensional feature fusion, and significantly improves the comprehensiveness and accuracy of the judgment of multiple violation behaviors such as missing the flag pole, walking and turning the wrist. Meanwhile, the application not only realizes automatic identification of violation behaviors, but also synchronously provides quantitative training indicators reflecting the rhythm control and body stability of athletes by fusing step frequency analysis and joint angle fluctuation analysis, so that the evaluation is deepened from qualitative judgment to quantitative analysis level, and direct parameter basis is provided for training improvement. In terms of practicability, the application adopts a lightweight model and a parallel processing architecture to ensure high recognition accuracy while meeting real-time monitoring requirements, and the system can be deployed relying on general hardware, thereby reducing the application threshold. In addition, the modular design enables the system to flexibly adapt to different numbers of cameras and adjustable judgment thresholds, thereby having good expansibility and being able to adapt to diversified training scenes and standard requirements.

[0095] The above merely provides the preferred embodiments of the present application, and the protection scope of the present application is not limited thereto, and any changes or substitutions within the technical scope disclosed by the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A multi-dimensional monitoring method for basketball dribbling around cones based on multiple perspectives, characterized in that, include: Deploy at least two cameras to simultaneously capture multiple video streams of the area where the pole moves; The multiple video streams are processed to detect key human points, basketballs, and marker poles in each video stream, and the detection results of the multiple video streams are fused to obtain the fused target state information. Based on the fused target state information, step frequency analysis and joint angle analysis are performed, and based on the analysis results and the spatial relationship between targets, it is determined whether there is any violation. The output includes multi-dimensional monitoring information, including violation determination results and motion analysis results.

2. The multi-dimensional monitoring method for basketball cone maneuvering based on multiple perspectives as described in claim 1, characterized in that, Deploying at least two cameras includes: distributing the cameras on both sides of the pole-movement path and ensuring that the field of view coverage areas of each camera overlap to eliminate blind spots; and using a timestamp alignment mechanism to synchronize the multiple video streams.

3. The multi-dimensional monitoring method for basketball cone-driving motion based on multiple perspectives as described in claim 1, characterized in that, Processing multiple video streams includes: using a cluster of lightweight deep learning models running in parallel to detect key points of human skeletons, the bounding box and center point of basketballs, and the bounding box and segmentation mask of marker poles from each video stream.

4. The multi-dimensional monitoring method for basketball cone-driving motion based on multiple perspectives as described in claim 3, characterized in that, The detection results of multiple video streams are merged as follows: the detection results of each video stream within the same time window are aligned; and a majority voting strategy is adopted, in which the detection results or violation judgments for the same target are consistent in more than a preset number of video streams, and are adopted as the merged target status information.

5. The multi-dimensional monitoring method for basketball cone maneuvering based on multiple perspectives as described in claim 1, characterized in that, Determining whether a violation exists includes: determining a missed turn violation based on the distance between the center point of the human body and the center point of the signpost; determining a signpost falling violation based on the tilt angle of the signpost's segmented mask; and comprehensively determining a walking violation by combining the cadence analysis results and joint angle analysis results.

6. The multi-dimensional monitoring method for basketball cone-driving motion based on multiple perspectives as described in claim 5, characterized in that, Step frequency analysis includes: extracting the vertical displacement time-series signal of key points of the human ankle, and performing frequency domain transformation on the vertical displacement time-series signal to extract the main frequency as the step frequency; joint angle analysis includes: calculating the knee joint angle and statistically analyzing the degree of temporal fluctuation of the knee joint angle.

7. The multi-dimensional monitoring method for basketball cone maneuvering based on multiple perspectives as described in claim 1, characterized in that, The output of multi-dimensional monitoring information includes: real-time rendering and display of the fused target status information, cadence change curve, joint angle fluctuation index and violation warning mark in the visualization interface; and asynchronous storage of violation judgment results and related motion analysis results.

8. A multi-dimensional monitoring system for basketball cone-driving motion based on multiple perspectives, characterized in that, For implementing the multi-dimensional monitoring method for basketball cone-driving motion based on any one of claims 1-7, the system comprises: The video acquisition synchronization module is used to deploy at least two cameras to simultaneously acquire multiple video streams from the area of ​​the pole-driving motion. The target detection fusion module is used to process the multiple video streams to detect human key points, basketballs and marker poles in each video stream, and fuse the detection results of the multiple video streams to obtain the fused target state information. The motion analysis and violation determination module is used to perform step frequency analysis and joint angle analysis based on the fused target state information, and to determine whether there is a violation based on the analysis results and the spatial relationship between targets. The multi-dimensional information output module is used to output multi-dimensional monitoring information, including violation judgment results and motion analysis results.

9. An electronic device, characterized in that, The electronic device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the multi-dimensional monitoring method for basketball dribbling around cones based on any one of claims 1-7.

10. A computer storage medium, characterized in that, The computer storage medium stores computer program instructions, which, when executed by a processor, implement the multi-dimensional monitoring method for basketball dribbling around cones based on any one of claims 1-7.

Images