Motion quality judgment system and method based on video analysis

Abstract

The invention relates to an action quality judgment system and method based on video analysis, and belongs to the field of physical training, and the system comprises at least one embedded high-definition camera which is used for capturing an exercise video of an exerciser in real time; the video processing unit is used for extracting key point information from the motion video; the action analysis module is used for performing action conformity analysis on the key point information; the quality judgment module is used for judging the action quality according to the action conformity analysis result; and the display module is used for displaying the action video, the analysis result and the judgment feedback. The system has the advantages that low-cost deployment is achieved, only the camera needs to be embedded in the system, additional sensors are not needed, and the deployment cost is reduced. Efficient video processing: adopting a vector database and an optimized video processing algorithm, and improving the accuracy and efficiency of key point extraction.

Description

Technical Field

[0001] The invention relates to an action quality evaluation system based on video analysis, and belongs to the field of physical training. Background Art

[0002] In traditional physical training, the evaluation of training results often relies on manual observation and recording. This method has problems such as strong subjectivity, low efficiency, and low accuracy. With the development of computer vision and edge computing technology, it has become possible to conduct training evaluation through automated systems. In existing technologies, some systems can capture human movements through cameras, but in physical training, the accuracy and standardization of movements are extremely high.

[0003] With the development of artificial intelligence technology, image analysis has been widely used in many fields. Especially in the field of sports training and physical training, real-time capture and evaluation of athletes' movements through video analysis technology has become an important means to improve training efficiency and quality. However, existing physical training evaluation systems often rely on complex sensor equipment and high costs, which limits their application in a wide range of occasions. Summary of the invention

[0004] In order to overcome the defects of the prior art, the present invention provides a motion quality evaluation system and method based on video analysis. The technical solution of the present invention is:

[0005] A motion quality evaluation system based on video analysis, comprising:

[0006] At least one built-in high-definition camera for capturing the sports video of the athlete in real time;

[0007] A video processing unit, used for extracting key point information from the motion video;

[0008] An action analysis module, used to perform action conformity analysis on the key point information; a quality evaluation module, used to evaluate the action quality according to the action conformity analysis result;

[0009] Display module, used to show action videos, analysis results and judges’ feedback.

[0010] The video processing unit comprises:

[0011] A pre-processing module for denoising, lighting correction, and resolution adjustment of the captured video;

[0012] The key point extraction module is used to identify the key parts of the athlete in the video and extract their position information.

[0013] The action analysis module comprises:

[0014] A deep learning module for learning key point trajectories and motion patterns of standard actions;

[0015] The conformity evaluation module is used to compare the extracted key point information with the learned standard action to determine the conformity of the action.

[0016] The quality evaluation module includes:

[0017] Scoring algorithm module, used to give scores based on action compliance;

[0018] Feedback generation module, used to generate improvement suggestions based on the scoring results.

[0019] A method for implementing the action quality evaluation system based on video analysis comprises the following steps:

[0020] (1) Use the built-in high-definition camera to capture the athlete's motion video;

[0021] (2) extracting key point information from the motion video;

[0022] (3) Conduct action conformity analysis on key point information;

[0023] (4) judging the action quality according to the action conformity analysis results;

[0024] (5) Display the action video, analysis results and judges’ feedback.

[0025] The step (2) is specifically as follows:

[0026] 2.1 When denoising the video, use the Gaussian filter algorithm to reduce the noise of the video image and improve the image quality. Among them, I(x,y): original image; I smooth (x,y): denoised image; G(i,j): Gaussian kernel; h: kernel size; a: standard deviation of the kernel;

[0027] 2.2 When performing illumination correction, apply illumination compensation algorithm to adjust the brightness and contrast of the image to adapt to different lighting conditions;

[0028] Among them, I out (x,y): corrected pixel value, i(x,y)): original pixel value; T(k): cumulative distribution function;

[0029] 2.3 When adjusting the resolution, use nearest neighbor interpolation to adjust the video resolution as needed to optimize the efficiency and quality of subsequent processing;

[0030] I res(x,y)=I(round(x / s x ),round(y / s y ));

[0031] 2.4 When performing key point detection, background subtraction is used to separate the mover from the background:

[0032]

[0033] F(x,y): foreground mask; It(x,y): current frame; B(x,y): background model; θ: threshold;

[0034] 2.5 Identify the key parts of the athlete in the video, such as joints, limb ends, etc.

[0035] R = det(M) - k(trace(M)) 2 , where M is the structure tensor; k is an empirical constant between 0.04 and 0.06; 2.6 uses the Lucas-Kanade method to track these key points in consecutive video frames, J T JΔx=J T b; J: Jacobian matrix; Δx: displacement vector; b: brightness change vector; 2.7 record the coordinate position of the key point, p t =(x t ,y t );pt:the position of the key point at time t;extract the velocity and acceleration dynamic information of the key point, vt: velocity at time t; Δt: time interval;

[0036] 2.8 Output the extracted key point information to the action analysis module for subsequent action compliance analysis, Output = {(pt, vt)}, Output: output key point information.

[0037] The step (3) is specifically as follows: 3.1 Use a deep learning model to learn the key point trajectory and motion pattern of the standard action, and use a recurrent neural network to learn the action pattern, h t =σ(W ih x t +b ih +W hh h t-1 +b hh );in,

[0038] ht: hidden state at time t;

[0039] xt: input at time t;

[0040] W ih ,b ih ,Whh ,b hh : weights and biases;

[0041] 3.2 Compare the extracted key point information with the learned standard action to determine the conformity of the action, and use the dynamic time warping algorithm to compare the time series data:

[0042] DTW distance: T: standard action sequence; S: actual action sequence; d: Euclidean distance; σ: sequence index arrangement; n: sequence length;

[0043] 3.3 Give scores based on the action compliance and design a scoring system based on error threshold. Score: score; C: action compliance; k: slope parameter; θ: threshold; 3.4 Generate improvement suggestions based on the scoring results;

[0044] 3.5 Output action compliance analysis results and improvement suggestions.

[0045] The step (4) is specifically as follows:

[0046] 4.1 Comprehensive score and action characteristics to give the final action quality judgment:

[0047] Quality=w1×Score+w2×Consistency+w3×Technique;

[0048] Quality: final action quality score; w1, w2, w3: weights of different scoring dimensions;

[0049] Consistency: action consistency score;

[0050] Technique: Technical execution score;

[0051] 4.2 Output action quality evaluation results and improvement suggestions.

[0052] A computer-readable storage medium is used to store a computer program for implementing the method.

[0053] A physical training all-in-one machine comprises the above-mentioned action quality evaluation system based on video analysis.

[0054] The advantages of the present invention are:

[0055] Low-cost deployment: The system only requires an embedded camera and no additional sensors, which reduces deployment costs.

[0056] Efficient video processing: The use of vector database and optimized video processing algorithms improves the accuracy and efficiency of key point extraction.

[0057] Deep learning: Using deep learning models for motion analysis improves the accuracy of motion compliance analysis.

[0058] Real-time feedback: The system can provide real-time evaluation results of movement quality, which helps to adjust the training plan in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a main structural block diagram of the system of the present invention. DETAILED DESCRIPTION

[0060] The present invention will be further described below in conjunction with specific embodiments, and the advantages and features of the present invention will become clearer as the description proceeds. However, these embodiments are exemplary only and do not constitute any limitation to the scope of the present invention. It should be understood by those skilled in the art that the details and forms of the technical solution of the present invention may be modified or replaced without departing from the spirit and scope of the present invention, but these modifications and replacements all fall within the scope of protection of the present invention.

[0061] See also Figure 1 The present invention relates to an action quality evaluation system based on video analysis, comprising:

[0062] At least one built-in high-definition camera 1, used to capture the sports video of the athlete in real time;

[0063] A video processing unit 2, used for extracting key point information from the motion video;

[0064] An action analysis module 3 is used to perform action conformity analysis on the key point information based on a deep learning algorithm;

[0065] A quality evaluation module 4 is used to evaluate the action quality according to the action conformity analysis result;

[0066] Display module 5 is used to display action videos, analysis results and evaluation feedback.

[0067] Based on the configuration of the above modules, the present invention achieves:

[0068] Efficiency: Ability to capture and analyze athletes’ movements in real time without waiting for manual scoring, thus significantly improving the efficiency of training and evaluation.

[0069] Accuracy: Using deep learning algorithms to quantify actions reduces the bias of human subjective judgment and improves the accuracy and consistency of judgment.

[0070] Objectivity: The system provides judgments based on video content and algorithm analysis results, rather than relying on personal experience or emotions, ensuring the objectivity of the judgments.

[0071] Real-time feedback: It can provide instant evaluation of movement quality and improvement suggestions, helping athletes to adjust their movements in time and optimize training effects.

[0072] The video processing unit comprises:

[0073] A pre-processing module for denoising, lighting correction, and resolution adjustment of the captured video;

[0074] The key point extraction module is used to identify the key parts of the athlete in the video and extract their position information.

[0075] The video preprocessing unit implements:

[0076] Enhanced Image Quality:

[0077] Denoising: By removing image noise, the accuracy of subsequent keypoint detection can be improved.

[0078] Lighting correction: Adapt to different lighting conditions to ensure accurate extraction of key point information in various environments.

[0079] Resolution adjustment: Optimize image resolution, balance processing speed and analysis quality, and adapt to computing devices with different performance.

[0080] Improve processing efficiency: By optimizing image quality and resolution, reduce computing resource consumption and increase processing speed.

[0081] robustness:

[0082] The design of the preprocessing module makes the system more adaptable to environmental changes (such as illumination changes) and ensures the robustness of the system.

[0083] Accurate key point positioning:

[0084] Key point extraction module: provides accurate input data for motion analysis by accurately identifying and extracting the key parts of the athlete in the video.

[0085] flexibility:

[0086] The preprocessing module can be adjusted according to different application scenarios and requirements to adapt to different video quality and environmental conditions.

[0087] The action analysis module comprises:

[0088] The deep learning module is used to learn the key point trajectory and motion pattern of the standard action; the conformity evaluation module is used to compare the extracted key point information with the learned standard action to determine the conformity of the action.

[0089] The motion analysis module implements:

[0090] High-precision motion recognition: The deep learning module is able to learn the complex patterns of standard movements through training, thereby recognizing and analyzing the movement trajectories and patterns of key points with high accuracy.

[0091] Powerful pattern recognition capabilities: Using the powerful feature extraction capabilities of deep neural networks, it is possible to identify and distinguish subtle differences in movements, improving the accuracy of movement analysis.

[0092] Adaptable: Able to adapt to different player sizes, speeds, and styles because deep learning models can learn from large amounts of diverse data.

[0093] The quality evaluation module includes:

[0094] Scoring algorithm module, used to give scores based on action compliance;

[0095] Feedback generation module, used to generate improvement suggestions based on the scoring results.

[0096] The above quality assessment module implements:

[0097] Standardized scoring: The scoring algorithm module can provide consistent scoring standards to ensure that the evaluation results of different individuals or at different times are comparable.

[0098] Instant Feedback: Improvement suggestions can be instantly generated based on the scoring results, helping athletes quickly understand their performance and make corresponding adjustments.

[0099] Personalized suggestions: The feedback generation module can provide customized feedback and suggestions based on the specific situation of each athlete, improving the relevance and effectiveness of training.

[0100] Motivational improvement: timely scoring and constructive feedback can motivate athletes and increase their training enthusiasm and participation.

[0101] The present invention also relates to a method for realizing the action quality evaluation system based on video analysis, comprising the following steps:

[0102] (1) Use the built-in high-definition camera to capture the athlete's motion video;

[0103] (2) extracting key point information from the motion video;

[0104] (3) Conduct action conformity analysis on key point information;

[0105] (4) judging the action quality according to the action conformity analysis results;

[0106] (5) Display the action video, analysis results and judges’ feedback.

[0107] The step (2) is specifically as follows:

[0108] 2.1 When denoising the video, use the Gaussian filter algorithm to reduce the noise of the video image and improve the image quality. Where, I(x,y): original image; Ismooth(x,y): denoised image; G(i,j): Gaussian kernel; h: kernel size; a: standard deviation of kernel;

[0109] 2.2 When performing illumination correction, apply illumination compensation algorithm to adjust the brightness and contrast of the image to adapt to different lighting conditions; Where, Iout(x,y): corrected pixel value, i(x,y): original pixel value; T(k): cumulative distribution function;

[0110] 2.3 When adjusting the resolution, use nearest neighbor interpolation to adjust the video resolution as needed to optimize the efficiency and quality of subsequent processing;

[0111] Ires(x,y)=I(round(x / sx),round(y / sy));

[0112] 2.4 When performing key point detection, background subtraction is used to separate the mover from the background:

[0113]

[0114] F(x,y): foreground mask; It(x,y): current frame; B(x,y): background model; θ: threshold;

[0115] 2.5 Identify the key parts of the athlete in the video, such as joints, limb ends, etc.

[0116] R = det(M) - k(trace(M))2, where M is the structure tensor; k is an empirical constant between 0.04 and 0.06;

[0117] 2.6 Using the Lucas-Kanade method, track these key points in consecutive video frames, J T JΔx=J T b; J: Jacobian matrix; Δx: displacement vector; b: brightness change vector; 2.7 record the coordinate position of the key point, p t =(x t ,y t );pt:the position of the key point at time t;extract the velocity and acceleration dynamic information of the key point, vt: velocity at time t; Δt: time interval;

[0118] 2.8 Output the extracted key point information to the action analysis module for subsequent action compliance analysis, Output = {(pt, vt)}, Output: output key point information.

[0119] The step (3) is specifically as follows: 3.1 Use a deep learning model to learn the key point trajectory and motion pattern of the standard action, and use a recurrent neural network to learn the action pattern, h t =σ(W ih x t +b ih +W hh h t-1 +b hh );in,

[0120] ht: hidden state at time t;

[0121] xt: input at time t;

[0122] Wih,bih,Whh,bhh: weights and biases;

[0123] 3.2 Compare the extracted key point information with the learned standard action to determine the conformity of the action, and use the dynamic time warping algorithm to compare the time series data:

[0124] DTW distance:

[0125] T: standard action sequence; S: actual action sequence; d: Euclidean distance; σ: sequence index arrangement; n: sequence length;

[0126] 3.3 Give scores based on the action compliance and design a scoring system based on error threshold. Score: score; C: action compliance; k: slope parameter; θ: threshold;

[0127] 3.4 Generate improvement suggestions based on the scoring results;

[0128] 3.5 Output action compliance analysis results and improvement suggestions.

[0129] The step (4) is specifically as follows:

[0130] 4.1 Comprehensive score and action characteristics to give the final action quality judgment:

[0131] Quality=w1×Score+w2×Consistency+w3×Technique;

[0132] Quality: final action quality score; w1, w2, w3: weights of different scoring dimensions;

[0133] ConsistencyConsistency: action consistency score;

[0134] TechniqueTechnique: Technical execution score;

[0135] 4.2 Output action quality evaluation results and improvement suggestions.

[0136] The present invention also relates to a computer-readable storage medium for storing a computer program for implementing the method.

[0137] The present invention also relates to an all-in-one physical training machine, comprising the above-mentioned action quality evaluation system based on video analysis.

[0138] In the present invention, the workflow and principle are as follows:

[0139] Video capture: Use the built-in high-definition camera to capture the athlete's motion video. This step is the basis of data collection and provides raw materials for subsequent analysis.

[0140] Video preprocessing: Denoising: Use Gaussian filtering algorithm to reduce the noise of video images and improve image quality.

[0141] Lighting Correction: Apply a lighting compensation algorithm to adjust the brightness and contrast of an image.

[0142] Resolution adjustment: Use the nearest neighbor interpolation method to adjust the video resolution and optimize the subsequent processing efficiency.

[0143] Key point detection: Use background subtraction to separate the subject from the background and extract key parts.

[0144] Key point extraction: Use algorithms such as Harris corner detection to identify key parts of the person in the video, such as joints and limb ends. Use the Lucas-Kanade method or other optical flow methods to track these key points in consecutive video frames. Record the coordinate positions of the key points and extract the velocity and acceleration dynamic information of the key points.

[0145] Action compliance analysis:

[0146] Deep learning models: Use deep learning models such as recurrent neural networks to learn key point trajectories and motion patterns of standard actions.

[0147] Conformity assessment: Use the dynamic time warping algorithm to compare the actual action sequence with the standard action sequence to determine the conformity of the action.

[0148] Scoring: Give scores based on the degree of action compliance and design a scoring system based on error thresholds. Feedback generation: Generate improvement suggestions based on the scoring results.

[0149] Action quality evaluation:

[0150] Comprehensive score: The final score of the movement quality is given by taking into account factors such as movement compliance, consistency and technical execution.

[0151] Output results: Output action quality evaluation results and improvement suggestions.

[0152] Results:

[0153] The display module shows action videos, analysis results and judgement feedback, providing an intuitive interactive interface.

[0154] The present invention uses automated video analysis technology to achieve real-time capture, analysis and evaluation of athletes' movements. Through preprocessing, key point extraction, movement conformity analysis and quality evaluation, the system can objectively and accurately evaluate the quality of movements and provide timely feedback and improvement suggestions. This method not only improves training efficiency, but also increases the scientificity and pertinence of training.

[0155] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A motion quality evaluation system based on video analysis, characterized in that: include: At least one built-in high-definition camera for capturing the sports video of the athlete in real time; A video processing unit, used for extracting key point information from the motion video; An action analysis module, used for performing action conformity analysis on the key point information; A quality evaluation module, used to evaluate the action quality according to the action conformity analysis result; Display module, used to show action videos, analysis results and judges’ feedback.

2. The system according to claim 1, characterized in that The video processing unit comprises: A pre-processing module for denoising, lighting correction, and resolution adjustment of the captured video; The key point extraction module is used to identify the key parts of the athlete in the video and extract their position information.

3. The system according to claim 1 or 2, characterized in that: The action analysis module comprises: A deep learning module for learning key point trajectories and motion patterns of standard actions; The conformity evaluation module is used to compare the extracted key point information with the learned standard action to determine the conformity of the action.

4. The system according to claim 3, characterized in that The quality evaluation module includes: Scoring algorithm module, used to give scores based on action compliance; Feedback generation module, used to generate improvement suggestions based on the scoring results.

5. A method for implementing the action quality evaluation system based on video analysis as described in any one of claims 1 to 4, characterized in that: The following steps are involved: (1) Use the built-in high-definition camera to capture the athlete's motion video; (2) extracting key point information from the motion video; (3) Conduct action conformity analysis on key point information; (4) judging the action quality according to the action conformity analysis results; (5) Display the action video, analysis results and judges’ feedback.

6. The method according to claim 5, characterized in that The step (2) is specifically as follows: 2.1 When denoising the video, use the Gaussian filter algorithm to reduce the noise of the video image and improve the image quality. Among them, I(x,y): original image; I smooth (x,y): denoised image; G(i,j): Gaussian kernel; h: kernel size; a: standard deviation of the kernel; 2.2 When performing illumination correction, apply illumination compensation algorithm to adjust the brightness and contrast of the image to adapt to different lighting conditions; Among them, I out (x,y): corrected pixel value, i(x,y)): original pixel value; T(k): cumulative distribution function; 2.3 When adjusting the resolution, use nearest neighbor interpolation to adjust the video resolution as needed to optimize the efficiency and quality of subsequent processing; I res (x,y)=I(round(x / s x ),round(y / s y )); 2.4 When performing key point detection, background subtraction is used to separate the mover from the background: F(x,y): foreground mask; It(x,y): current frame; B(x,y): background model; θ: threshold; 2.5 Identify the key parts of the athlete in the video; R = det(M) - k(trace(M)) 2 , where M is the structural tensor; k is an empirical constant, between 0.04 and 0.06; 2.6 Using the Lucas-Kanade method, track these key points in consecutive video frames, J T JΔx=J T b; J: Jacobian matrix; Δx: displacement vector; b: brightness change vector; 2.7 Record the coordinates of the key points, p t =(x t ,y t );pt:the position of the key point at time t;extract the velocity and acceleration dynamic information of the key point, vt: velocity at time t; Δt: time interval; 2.8 Output the extracted key point information to the action analysis module for subsequent action compliance analysis, Output = {(pt, vt)}, Output: output key point information.

7. The method according to claim 6, characterized in that The step (3) is specifically as follows: 3.1 Use deep learning models to learn key point trajectories and motion patterns of standard actions, and use recurrent neural networks to learn action patterns. t =σ(W ih x t +b ih +W hh h t-1 +b hh ); in, ht: hidden state at time t; xt: input at time t; W ih ,b ih ,W hh ,b hh : weights and biases; 3.2 Compare the extracted key point information with the learned standard action to determine the conformity of the action, and use the dynamic time warping algorithm to compare the time series data: DTW distance: T: standard action sequence; S: actual action sequence; d: Euclidean distance; σ: sequence index arrangement; n: sequence length; 3.3 Give scores based on the action compliance and design a scoring system based on error threshold. Score: score; C: Action compliance; k: slope parameter; θ: threshold; 3.4 Generate improvement suggestions based on the scoring results; 3.5 Output action compliance analysis results and improvement suggestions.

8. The method according to claim 6, characterized in that The step (4) is specifically as follows: 4.1 Comprehensive score and action characteristics to give the final action quality judgment: Quality=w1×Score+w2×Consistency+w3×Technique; Quality: final action quality score; w1, w2, w3: weights of different scoring dimensions; Consistency: action consistency score; Technique: Technical execution score; 4.2 Output action quality evaluation results and improvement suggestions.

9. A computer-readable storage medium, characterized in that: Used to store a computer program for implementing the method according to any one of claims 5 to 8.

10. A physical training all-in-one machine, characterized in that: The invention comprises the action quality evaluation system based on video analysis as described in any one of claims 1 to 4.

Images