Taekwondo action posture detection and analysis method and system based on visual identification
Through visual recognition technology, keyframes of Taekwondo movements are extracted and feature data is integrated to build a detection model, which solves the problem of insufficient Taekwondo movement posture recognition ability in the existing technology, and achieves more accurate and efficient motion analysis.
Patent Information
- Application Number
- CN202410995094.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing visual recognition technology is difficult to effectively identify and analyze Taekwondo movement postures, resulting in insufficient ability to recognize and analyze Taekwondo movements.
Using the Taekwondo action posture detection and analysis method based on visual recognition, video data is collected through the camera module, preprocessing and segmentation processing, keyframes are extracted, and key points, angles, motion trajectories and shape features are extracted for each keyframe. Fusion feature data is used to build an action posture detection model for detection.
It improves the ability to identify and analyze Taekwondo movements, achieves more accurate and efficient movement posture detection, and is suitable for athlete movement evaluation and training optimization, referee judgment and other fields.
Smart Images

Figure CN118942006A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of visual recognition technology, and in particular to a Taekwondo action posture detection and analysis method and system based on visual recognition. Background Art
[0002] In recent years, with the rapid development of computer vision technology, people are increasingly applying it to various fields. Among them, athlete action recognition is a field where computer vision technology is widely used. Athlete action recognition based on computer vision technology can help athletes evaluate their actions and optimize their training in the field of sports science. In the field of e-sports, athlete action recognition can help referees determine whether athletes have committed fouls, thereby ensuring the fairness of the game. Therefore, how to detect and analyze Taekwondo action postures based on visual recognition and improve the ability to recognize and analyze Taekwondo actions has become an urgent problem to be solved. Summary of the invention
[0003] The main purpose of this application is to provide a Taekwondo action posture detection and analysis method and system based on visual recognition, aiming to improve the ability to recognize and analyze Taekwondo actions.
[0004] In order to achieve the above-mentioned invention object, the present application proposes a Taekwondo action posture detection and analysis method based on visual recognition, comprising:
[0005] A Taekwondo action posture detection and analysis method based on visual recognition, the method comprising:
[0006] The camera module collects video data of the Taekwondo action;
[0007] Preprocessing the video data to obtain first video data;
[0008] Segmenting the first video data to extract all key frames of the Taekwondo action;
[0009] Extract key point features from each key frame to obtain position feature data, angle feature data, motion trajectory feature data and shape feature data;
[0010] Fusing the position feature data, the angle feature data, the motion trajectory feature data and the shape feature data to obtain a multi-feature fusion feature vector;
[0011] An action posture detection model is constructed, and the multi-feature fusion feature vector is input into the action posture recognition model to detect Taekwondo action postures, so as to obtain Taekwondo action posture detection results.
[0012] Further, the first video data is segmented to extract all key frames of the Taekwondo action, specifically including:
[0013] dividing the first video data into a plurality of video segments based on a preset time interval;
[0014] Convert each of the video clips into a video frame based on preset conversion parameters;
[0015] The key action postures of the video frames are extracted based on a preset key action posture sequence to obtain key frames of Taekwondo actions.
[0016] Furthermore, key point feature extraction is performed on each of the key frames to obtain position feature data, specifically including:
[0017] Preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale conversion, noise reduction, and scale normalization;
[0018] Identify the human skeleton structure of each of the first key frames to obtain key point data of the human body, wherein the key points of the human skeleton structure include the head, shoulders, elbows, wrists, hips, knees and ankles;
[0019] Calculate the coordinates of each key point according to the key point data to obtain the position data of each key point;
[0020] The position data of the key points are subjected to denoising and smoothing processing to obtain position feature data.
[0021] Furthermore, key point feature extraction is performed on each of the key frames to obtain angle feature data, specifically including:
[0022] Extract key points from each of the key frames based on the corner point detection to obtain key points from each of the key frames;
[0023] Calculate a corresponding key point descriptor according to each of the key points;
[0024] The surrounding angle features of each key point are calculated according to the key point descriptor to obtain angle feature data.
[0025] Furthermore, key point feature extraction is performed on each of the key frames to obtain motion trajectory feature data, specifically including:
[0026] Preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale conversion, noise reduction, and scale normalization;
[0027] Detect key points of each of the first key frames based on a scale-invariant feature transformation algorithm, and extract feature descriptors of each of the key points;
[0028] Performing feature matching between each adjacent first key frame according to the feature descriptor of each key point to obtain corresponding matching key points;
[0029] The motion trajectory of the corresponding matched key points is calculated based on the optical flow method to obtain motion trajectory feature data.
[0030] Furthermore, key point feature extraction is performed on each of the key frames to obtain shape feature data, specifically including:
[0031] Preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale conversion, noise reduction, and scale normalization;
[0032] Grayscale each of the first key frames to obtain a grayscale key frame;
[0033] Convert each of the grayscale key frames into a grayscale image based on a programming language;
[0034] Performing threshold segmentation on the grayscale image to obtain a binary image;
[0035] Performing contour detection on the binary image based on an edge detection algorithm;
[0036] Shape feature extraction is performed on the contour to obtain shape feature data, wherein the shape feature data includes but is not limited to contour features of Taekwondo action postures.
[0037] Furthermore, before the step of constructing the action posture detection model, it specifically includes:
[0038] Acquire a multi-feature fusion feature vector data set, and divide the multi-feature fusion feature vector data set into a multi-feature fusion feature vector training data set and a multi-feature fusion feature vector test data set;
[0039] The motion posture detection model is trained based on the multi-feature fusion feature vector training data set, and the motion posture detection model is tested based on the multi-feature fusion feature vector test data set until the prediction accuracy of the motion posture detection model meets a preset accuracy threshold, and the training of the motion posture detection model is stopped.
[0040] A Taekwondo action posture detection and analysis system based on visual recognition, the system comprising:
[0041] A video data module, used for collecting video data of the Taekwondo movements based on a camera module;
[0042] A preprocessing module, used for preprocessing the video data to obtain first video data;
[0043] A key frame extraction module, used for segmenting the first video data and extracting all key frames of the Taekwondo action;
[0044] A feature extraction module is used to extract key point features from each key frame to obtain position feature data, angle feature data, motion trajectory feature data and shape feature data;
[0045] A multi-feature fusion feature vector module is used to fuse the position feature data, the angle feature data, the motion trajectory feature data and the shape feature data to obtain a multi-feature fusion feature vector;
[0046] The detection result module is used to construct an action posture detection model, input the multi-feature fusion feature vector into the action posture recognition model to detect Taekwondo action postures, and obtain Taekwondo action posture detection results.
[0047] The present application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of any one of the above-mentioned Taekwondo action posture detection and analysis methods based on visual recognition are implemented.
[0048] The present application also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of any one of the above-mentioned Taekwondo action posture detection and analysis methods based on visual recognition are implemented.
[0049] The present application relates to the field of visual recognition technology, and in particular to a method and system for detecting and analyzing Taekwondo action postures based on visual recognition, the method comprising: a camera module collects video data of Taekwondo actions; pre-processes the video data to obtain first video data; segments the first video data to extract all key frames of Taekwondo actions; extracts key point features from each key frame to obtain position feature data, angle feature data, motion trajectory feature data, and shape feature data; fuses the position feature data, angle feature data, motion trajectory feature data, and shape feature data to obtain a multi-feature fusion feature vector; constructs an action posture detection model, and inputs the multi-feature fusion feature vector into the action posture recognition model to detect Taekwondo action postures, and obtains Taekwondo action posture detection results. Through the above method, the ability to identify and analyze Taekwondo actions is improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] Figure 1This is a flow chart of an embodiment of a method for detecting and analyzing Taekwondo movements and postures based on visual recognition in this application;
[0051] Figure 2 This is a schematic diagram of the structure of an embodiment of a Taekwondo action posture detection and analysis system based on visual recognition in this application;
[0052] Figure 3 This is a schematic block diagram of the structure of an embodiment of a computer device of the present application.
[0053] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0054] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0055] Reference Figure 1 The embodiment of the present application provides a Taekwondo action posture detection and analysis method based on visual recognition, including steps S10-S60. The detailed description of each step of the Taekwondo action posture detection and analysis method based on visual recognition is as follows.
[0056] S10, collecting video data of the Taekwondo action based on a camera module;
[0057] In this embodiment, the video data needs to be decoded into raw pixel data during the preprocessing process, by real-time acquisition by a camera or reading an existing video file from a storage device. Individual frames are extracted from the video, which will be used for subsequent analysis or processing. A sample is extracted every few frames within a specific time interval. When processing the video, it may be necessary to adjust the frame size to a fixed size. Each frame is standardized or normalized. This usually involves scaling the pixel values to a range of 0 to 1 or -1 to 1, which helps to improve efficiency in subsequent model training or analysis.
[0058] S20, preprocessing the video data to obtain first video data;
[0059] In this embodiment, specifically, in one embodiment, real-time acquisition is performed by a camera, or an existing video file is read from a storage device. During the preprocessing process, the video data needs to be decoded into raw pixel data. Individual frames are extracted from the video, which will be used for subsequent analysis or processing. According to extracting a sample every few frames within a specific time interval, when processing the video, it may be necessary to adjust the size of the frame to a fixed size. Each frame is standardized or normalized. This usually involves scaling the pixel values to a range of 0 to 1 or -1 to 1, which helps to improve efficiency in subsequent model training or analysis.
[0060] S30, segmenting the first video data to extract all key frames of the Taekwondo action;
[0061] In this embodiment, the first video data is segmented and processed to extract all key frames of the Taekwondo movements, including: segmenting the first video data into multiple video segments based on a preset time interval; converting each of the video segments into a video frame based on preset conversion parameters; and extracting key action postures from the video frames based on a preset key action posture sequence to obtain key frames of the Taekwondo movements.
[0062] S40, extracting key point features from each of the key frames to obtain position feature data, angle feature data, motion trajectory feature data and shape feature data.
[0063] In this embodiment, key point features are extracted from each of the key frames to obtain position feature data, including: preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale, noise reduction and scale normalization; identifying the human skeleton structure of each of the first key frames to obtain key point data of the human body, wherein the key points of the human skeleton structure include the head, shoulders, elbows, wrists, hips, knees and ankles; and calculating the coordinates of each of the key points based on the key point data to obtain each of the key points.
[0064] Key point features are extracted for each of the key frames to obtain angle feature data, including: extracting key points for each of the key frames based on the corner point detection to obtain key points for each of the key frames; calculating corresponding key point descriptors according to each of the key points; and calculating surrounding angle features of each of the key points according to the key point descriptors to obtain angle feature data.
[0065] Key point features are extracted from each of the key frames to obtain motion trajectory feature data, including: preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale, noise reduction and scale normalization; detecting key points of each of the first key frames based on a scale-invariant feature transformation algorithm to extract a feature descriptor of each of the key points; performing feature matching between each adjacent first key frames based on the feature descriptor of each of the key points to obtain corresponding matching key points; and calculating motion trajectories of the corresponding matching key points based on an optical flow method to obtain motion trajectory feature data.
[0066] Key point features are extracted from each of the key frames to obtain shape feature data, including: preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to graying, noise reduction and scale normalization; graying each of the first key frames to obtain a grayed key frame; converting each of the grayed key frames into a grayed image based on a programming language; performing threshold segmentation on the grayed image to obtain a binary image; performing contour detection on the binary image based on an edge detection algorithm; performing shape feature extraction on the contour to obtain shape feature data, wherein the shape feature data includes but is not limited to contour features of Taekwondo action postures.
[0067] S50, fusing the position feature data, the angle feature data, the motion trajectory feature data and the shape feature data to obtain a multi-feature fusion feature vector;
[0068] In this embodiment, based on a convolutional neural network, the position feature data, the angle feature data, the motion trajectory feature data and the shape feature data are respectively converted into a position feature vector, an angle feature vector, a motion trajectory feature vector and a shape feature vector; weights are respectively assigned to the position feature vector, the angle feature vector, the motion trajectory feature vector and the shape feature vector to obtain corresponding weight coefficients; the position feature vector, the angle feature vector, the motion trajectory feature vector and the shape feature vector are respectively multiplied by their corresponding weight coefficients to obtain corresponding weighted feature vectors; the weighted feature vectors corresponding to the position feature vector, the angle feature vector, the motion trajectory feature vector and the shape feature vector are linearly added to obtain a multi-feature fusion vector.
[0069] S60, constructing an action posture detection model, inputting the multi-feature fusion feature vector into the action posture recognition model to detect Taekwondo action postures, and obtaining Taekwondo action posture detection results.
[0070] In the present embodiment, in one embodiment, a video data set of action postures is annotated, and the action postures in each video need to be annotated. The features include position feature data, angle feature data, motion trajectory feature data, and shape feature data, etc. The feature data, angle feature data, motion trajectory feature data, and shape feature data are combined to form a comprehensive feature vector. A motion posture detection model is constructed based on a convolutional neural network (CNN) and a recurrent neural network (RNN) based on deep learning. These models can learn the temporal and spatial relationship of actions from feature vectors, thereby achieving efficient action recognition and posture detection. The model is trained and optimized using the prepared data set and feature vector. During the training process, cross-validation or other methods are used to ensure the generalization ability and accuracy of the model. During the optimization process, the hyperparameters such as the structure of the model, the loss function, and the learning rate are adjusted to improve the performance and stability of the model.
[0071] After the model training is completed, the model is used to detect postures in new Taekwondo action videos. Input video frames or frame sequences, and the model will output the predicted posture category or action trajectory information for each frame or each action clip. The detection result is a labeled action category, or the athlete's specific action trajectory or key point position sequence.
[0072] In one embodiment, the first video data is segmented and processed to extract all key frames of the Taekwondo action, specifically including: segmenting the first video data into multiple video segments based on a preset time interval; converting each of the video segments into a video frame based on preset conversion parameters; extracting key action postures from the video frames based on a preset key action posture sequence to obtain key frames of the Taekwondo action.
[0073] In this embodiment, in one embodiment, the original first video data is divided into a plurality of video clips at a preset time interval. For example, if the preset time interval is 2 seconds and the total length of the video is 10 seconds, the video will be divided into 5 2-second video clips. Set the time interval, in this embodiment, set the time interval to 2 seconds, loop through every 2 seconds of the video, divide the video clips, and then, according to the preset conversion parameters, convert each video clip into a video frame. For example, the preset conversion parameters may be to extract 30 frames per second. Load the video clip, according to the frame rate parameter, in this embodiment, set the frame rate parameter to 30 frames per second to extract the frame, and finally, extract the key action posture of the extracted video frame based on the preset key action posture sequence to obtain the key frame of the Taekwondo action. In this embodiment, the key actions include kicking, jumping, rotating and other actions. Use computer vision and action recognition algorithms to identify and extract the action postures in each frame. According to the preset key action posture sequence, extract the frames containing these key actions from all frames as key frames.
[0074] In one embodiment, key point feature extraction is performed on each of the key frames to obtain position feature data, specifically including: preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale, noise reduction and scale normalization; identifying the human skeleton structure of each of the first key frames to obtain key point data of the human body, wherein the key points of the human skeleton structure include the head, shoulders, elbows, wrists, hips, knees and ankles; calculating the coordinates of each of the key points according to the key point data to obtain the position data of each of the key points; and denoising and smoothing the position data of the key points to obtain position feature data.
[0075] In this embodiment, specifically, each key frame is preprocessed, and the preprocessing steps include grayscale, denoising and scale normalization. Grayscale is to convert a color image into a grayscale image, denoising is to use a filter to reduce noise in the image, and scale normalization is to adjust the image size so that the input image has a consistent size.
[0076] Use the pre-trained model to identify the human skeleton structure and obtain the data of each key point. Use the deep learning model to identify and extract key points. Key points include the head, shoulders, elbows, wrists, hips, knees, and ankles. Calculate the coordinates of each key point based on the key point data. Each key point is represented by a set of (x, y) coordinates and may also have a confidence value. Extract the coordinate data of all key points. De-noise and smooth the position data of the key points to obtain more stable position information. Use a smoothing algorithm to reduce the noise of the position data.
[0077] In one embodiment, key point features are extracted for each of the key frames to obtain angle feature data, specifically including: extracting key points for each of the key frames based on the corner point detection to obtain key points for each of the key frames; calculating a corresponding key point descriptor according to each of the key points; and calculating surrounding angle features of each of the key points according to the key point descriptors to obtain angle feature data.
[0078] In this embodiment, in one embodiment, there is a video recording a Taekwondo player performing a series of actions, and key point features are extracted for each key frame to obtain angle feature data in order to analyze the player's action posture. First, multiple key frames are extracted from the video. These key frames are frames in the video that contain important action changes or postures. Specifically, key frame 1: the player's preparation action posture, key frame 2: the player starts to kick, and key frame 3: the highest point of the player's kick.
[0079] Keypoints are extracted for each keyframe. These keypoints are points with significant features in the image. For keyframe 1, some keypoints are detected at the shoulder, elbow, knee and ankle of the player. For keyframe 2, some keypoints are detected at the shoulder, elbow, knee and ankle of the kicking leg. For keyframe 3, some keypoints are detected at the shoulder, elbow, knee and raised ankle of the player. For each keypoint, a descriptor is calculated. The descriptor is a set of feature vectors that describe the characteristics of the image area around the keypoint.
[0080] Specifically, for a key point at the player's shoulder position in key frame 1, the change in pixel grayscale value within a certain radius around it is calculated to generate a descriptor. Based on each key point descriptor, the surrounding angle feature of each key point is calculated. The angle feature describes the direction of the image structure around the key point.
[0081] In one embodiment, key point feature extraction is performed on each of the key frames to obtain motion trajectory feature data, which specifically includes: preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale, noise reduction and scale normalization; detecting key points of each of the first key frames based on a scale-invariant feature transformation algorithm to extract a feature descriptor of each of the key points; performing feature matching between each adjacent first key frames based on the feature descriptor of each of the key points to obtain corresponding matching key points; and calculating motion trajectories of the corresponding matching key points based on an optical flow method to obtain motion trajectory feature data.
[0082] In this embodiment, there are three key frames in the video recording a kicking action: Key frame 1: the player is in preparation, the shoulder key point position is (100, 200), key frame 2: the player starts to kick, and the shoulder key point position is (105, 205). Key frame 3: the player's kick reaches the highest point, and the shoulder key point position is (110, 210). Key frames 1, 2, and 3 are converted to grayscale images, and a Gaussian filter is applied to reduce noise and normalize the image scale. The S IFT algorithm is used to detect the key points in each key frame and extract feature descriptors. The shoulder key point feature descriptor of key frame 1 is matched with the shoulder key point feature descriptor of key frame 2. The shoulder key point feature descriptor of key frame 2 is matched with the shoulder key point feature descriptor of key frame 3. The optical flow method is used to calculate the motion trajectory of the shoulder key point between adjacent frames, from key frame 1 (100, 200) to key frame 2 (105, 205), and from key frame 2 (105, 205) to key frame 3 (110, 210). The motion trajectory feature data of the shoulder key point is [(100, 200), (105, 205), (110, 210)]. Through these motion trajectory feature data, the player's movements are analyzed in detail, thereby achieving the purpose of Taekwondo action posture detection and analysis.
[0083] In one embodiment, key point features are extracted from each of the key frames to obtain shape feature data, which specifically includes: preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale, noise reduction and scale normalization; grayscale each of the first key frames to obtain a grayscale key frame; converting each of the grayscale key frames into a grayscale image based on a programming language; threshold segmenting the grayscale image to obtain a binary image; performing contour detection on the binary image based on an edge detection algorithm; performing shape feature extraction on the contour to obtain shape feature data, wherein the shape feature data includes but is not limited to contour features of Taekwondo action postures.
[0084] In this embodiment, specifically, in one embodiment, there are three key frames in the video recording a kicking action, key frame 1: the player's preparation posture, the detected outline is the outline of a standing person. Key frame 2: the player starts to kick, and the detected outline is the outline of a person with one leg raised. Key frame 3: the player's kick reaches the highest point, and the detected outline is the outline of a person with one leg fully raised. Key frames 1, 2, and 3 are converted into grayscale images, and a Gaussian filter is applied to reduce noise and normalize the image scale. Using a programming language, in this embodiment, Python language is used to convert each grayscale key frame into a grayscale image. Threshold segmentation is performed on the grayscale image to obtain a binary image. The Canny algorithm is used for edge detection, and then contour detection is performed.
[0085] The shape feature of the detected contour is extracted, and the area of the contour is calculated as the shape feature. Finally, the shape feature data of key frame 1 is extracted as the contour area of the standing posture, the shape feature data of key frame 2 is the contour area of the posture of one leg raised, and the shape feature data of key frame 3 is the contour area of the posture of one leg fully raised. Through these shape feature data, the player's movements are analyzed in detail, thereby achieving the purpose of Taekwondo movement posture detection and analysis.
[0086] In one embodiment, before the step of constructing the motion posture detection model, it specifically includes: obtaining a multi-feature fusion feature vector data set, dividing the multi-feature fusion feature vector data set into a multi-feature fusion feature vector training data set and a multi-feature fusion feature vector test data set; training the motion posture detection model based on the multi-feature fusion feature vector training data set, and testing the motion posture detection model based on the multi-feature fusion feature vector test data set, until the prediction accuracy of the motion posture detection model meets a preset accuracy threshold, and stopping training the motion posture detection model.
[0087] In this embodiment, specifically, in one embodiment, a large number of videos of Taekwondo players performing various actions are collected, and the following features are extracted through the aforementioned steps: Shape features: such as the area and circumference of the leg contour. Movement features: such as the speed and acceleration of the leg lifting. Angle features: such as the angles of the knee joint and hip joint. These features are integrated into a multi-feature fusion feature vector data set, and the data set is divided into a training data set and a test data set.
[0088] During the training process, the training data set is trained using the random forest algorithm, and the model parameters are continuously adjusted until the model's prediction accuracy on the test data set reaches 90%. In this way, we can build a highly accurate Taekwondo action posture detection model for practical applications such as player action analysis, training assistance, etc.
[0089] Reference Figure 2 The present application provides a Taekwondo action posture detection and analysis system based on visual recognition, the system comprising:
[0090] A video data module 10, used for collecting video data of the Taekwondo movements based on a camera module;
[0091] A preprocessing module 20, configured to preprocess the video data to obtain first video data;
[0092] A key frame extraction module 30 is used to segment the first video data and extract all key frames of the Taekwondo action;
[0093] A feature extraction module 40 is used to extract key point features from each key frame to obtain position feature data, angle feature data, motion trajectory feature data and shape feature data;
[0094] A multi-feature fusion feature vector module 50 is used to fuse the position feature data, the angle feature data, the motion trajectory feature data and the shape feature data to obtain a multi-feature fusion feature vector;
[0095] The detection result module 60 is used to construct an action posture detection model, input the multi-feature fusion feature vector into the action posture recognition model to detect Taekwondo action postures, and obtain Taekwondo action posture detection results.
[0096] As described above, it can be understood that the various components of the Taekwondo action posture detection and analysis system based on visual recognition proposed in this application can realize the functions of any one of the Taekwondo action posture detection and analysis methods based on visual recognition as described above.
[0097] In one embodiment, the key frame extraction module 300 further includes executing:
[0098] dividing the first video data into a plurality of video segments based on a preset time interval;
[0099] Convert each of the video clips into a video frame based on preset conversion parameters;
[0100] The key action postures of the video frames are extracted based on a preset key action posture sequence to obtain key frames of Taekwondo actions.
[0101] In one embodiment, the feature extraction module 40 further includes executing:
[0102] Preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale conversion, noise reduction, and scale normalization;
[0103] Identify the human skeleton structure of each of the first key frames to obtain key point data of the human body, wherein the key points of the human skeleton structure include the head, shoulders, elbows, wrists, hips, knees and ankles;
[0104] Calculate the coordinates of each key point according to the key point data to obtain the position data of each key point;
[0105] The position data of the key points are subjected to denoising and smoothing processing to obtain position feature data.
[0106] In one embodiment, the feature extraction module 40 further includes executing:
[0107] Extract key points from each of the key frames based on the corner point detection to obtain key points from each of the key frames;
[0108] Calculate a corresponding key point descriptor according to each of the key points;
[0109] The surrounding angle features of each key point are calculated according to the key point descriptor to obtain angle feature data.
[0110] In one embodiment, the feature extraction module 40 further includes executing:
[0111] Preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale conversion, noise reduction, and scale normalization;
[0112] Detect key points of each of the first key frames based on a scale-invariant feature transformation algorithm, and extract feature descriptors of each of the key points;
[0113] Performing feature matching between each adjacent first key frame according to the feature descriptor of each key point to obtain corresponding matching key points;
[0114] The motion trajectory of the corresponding matched key points is calculated based on the optical flow method to obtain motion trajectory feature data.
[0115] In one embodiment, the feature extraction module 40 further includes executing:
[0116] Preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale conversion, noise reduction, and scale normalization;
[0117] Grayscale each of the first key frames to obtain a grayscale key frame;
[0118] Convert each of the grayscale key frames into a grayscale image based on a programming language;
[0119] Performing threshold segmentation on the grayscale image to obtain a binary image;
[0120] Performing contour detection on the binary image based on an edge detection algorithm;
[0121] Shape feature extraction is performed on the contour to obtain shape feature data, wherein the shape feature data includes but is not limited to contour features of Taekwondo action postures.
[0122] In one embodiment, the detection result module 60 further includes executing:
[0123] Acquire a multi-feature fusion feature vector data set, and divide the multi-feature fusion feature vector data set into a multi-feature fusion feature vector training data set and a multi-feature fusion feature vector test data set;
[0124] The motion posture detection model is trained based on the multi-feature fusion feature vector training data set, and the motion posture detection model is tested based on the multi-feature fusion feature vector test data set until the prediction accuracy of the motion posture detection model meets a preset accuracy threshold, and the training of the motion posture detection model is stopped.
[0125] Reference Figure 3 In an embodiment of the present application, a computer device is also provided, the internal structure of which can be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a network interface, a display device and an input device connected through a system bus. Among them, the network interface of the computer device is used to communicate with an external terminal through a network connection. The display device of the computer device is used to display an interactive page. The input device of the computer device is used to receive user input. The processor designed for the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium. The non-volatile storage medium stores an operating system, a computer program and a database. The database of the computer device is used to store raw data. When the computer program is executed by the processor, a Taekwondo action posture detection and analysis method based on visual recognition is implemented.
[0126] The processor executes the above-mentioned method for detecting and analyzing Taekwondo action postures based on visual recognition, which includes: the camera module collects video data of Taekwondo action; pre-processes the video data to obtain the first video data; segments the first video data to extract all key frames of Taekwondo action; extracts key point features of each key frame to obtain position feature data, angle feature data, motion trajectory feature data and shape feature data; fuses the position feature data, angle feature data, motion trajectory feature data and shape feature data to obtain a multi-feature fusion feature vector; constructs an action posture detection model, and inputs the multi-feature fusion feature vector into the action posture recognition model to detect Taekwondo action postures, and obtains Taekwondo action posture detection results. Through the above method, the ability to recognize and analyze Taekwondo actions is improved.
[0127] The present application also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by the processor, a Taekwondo action posture detection and analysis method based on visual recognition is implemented, wherein a camera module collects video data of Taekwondo actions; the video data is preprocessed to obtain first video data; the first video data is segmented to extract all key frames of the Taekwondo actions; key point features are extracted for each key frame to obtain position feature data, angle feature data, motion trajectory feature data and shape feature data; the position feature data, angle feature data, motion trajectory feature data and shape feature data are fused to obtain a multi-feature fusion feature vector; an action posture detection model is constructed, and the multi-feature fusion feature vector is input into an action posture recognition model to detect Taekwondo action postures, thereby obtaining Taekwondo action posture detection results.
[0128] The computer-readable storage medium provides a method for detecting and analyzing Taekwondo action postures based on visual recognition, the method comprising: a camera module collects video data of Taekwondo actions; preprocesses the video data to obtain first video data; segments the first video data to extract all key frames of Taekwondo actions; extracts key point features of each key frame to obtain position feature data, angle feature data, motion trajectory feature data and shape feature data; fuses the position feature data, angle feature data, motion trajectory feature data and shape feature data to obtain a multi-feature fusion feature vector; constructs an action posture detection model, inputs the multi-feature fusion feature vector into the action posture recognition model to detect Taekwondo action postures, and obtains Taekwondo action posture detection results. Through the above method, the ability to identify and analyze Taekwondo actions is improved.
[0129] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0130] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the existence of other identical elements in the process, device, article or method including the element.
[0131] The above description is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.
Claims
1. A Taekwondo action posture detection and analysis method based on visual recognition, characterized in that: include: Collecting video data of the Taekwondo movements based on a camera module; Preprocessing the video data to obtain first video data; Segmenting the first video data to extract all key frames of the Taekwondo action; Extract key point features from each key frame to obtain position feature data, angle feature data, motion trajectory feature data and shape feature data; Fusing the position feature data, the angle feature data, the motion trajectory feature data and the shape feature data to obtain a multi-feature fusion feature vector; An action posture detection model is constructed, and the multi-feature fusion feature vector is input into the action posture recognition model to detect Taekwondo action postures, so as to obtain Taekwondo action posture detection results.
2. The method for detecting and analyzing Taekwondo movements and postures based on visual recognition as claimed in claim 1, characterized in that: The first video data is segmented to extract all key frames of the Taekwondo action, specifically including: dividing the first video data into a plurality of video segments based on a preset time interval; Convert each of the video clips into a video frame based on preset conversion parameters; The key action postures of the video frames are extracted based on a preset key action posture sequence to obtain key frames of Taekwondo actions.
3. The method for detecting and analyzing Taekwondo movements and postures based on visual recognition as claimed in claim 1, characterized in that: Key point feature extraction is performed on each of the key frames to obtain position feature data, specifically including: Preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale conversion, noise reduction, and scale normalization; Identify the human skeleton structure of each of the first key frames to obtain key point data of the human body, wherein the key points of the human skeleton structure include the head, shoulders, elbows, wrists, hips, knees and ankles; Calculate the coordinates of each key point according to the key point data to obtain the position data of each key point; The position data of the key points are subjected to denoising and smoothing processing to obtain position feature data.
4. The method for detecting and analyzing Taekwondo movements and postures based on visual recognition as claimed in claim 1, characterized in that: Key point feature extraction is performed on each of the key frames to obtain angle feature data, specifically including: Extract key points from each of the key frames based on the corner point detection to obtain key points from each of the key frames; Calculate a corresponding key point descriptor according to each of the key points; The surrounding angle features of each key point are calculated according to the key point descriptor to obtain angle feature data.
5. The method for detecting and analyzing Taekwondo movements and postures based on visual recognition as claimed in claim 1, characterized in that: Extract key point features from each key frame to obtain motion trajectory feature data, specifically including: Preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale conversion, noise reduction, and scale normalization; Detect key points of each of the first key frames based on a scale-invariant feature transformation algorithm, and extract feature descriptors of each of the key points; Performing feature matching between each adjacent first key frame according to the feature descriptor of each key point to obtain corresponding matching key points; The motion trajectory of the corresponding matched key points is calculated based on the optical flow method to obtain motion trajectory feature data.
6. The method for detecting and analyzing Taekwondo movements and postures based on visual recognition as claimed in claim 1, characterized in that: Key point feature extraction is performed on each of the key frames to obtain shape feature data, specifically including: Preprocessing each of the key frames to obtain a first key frame, wherein the preprocessing includes but is not limited to grayscale conversion, noise reduction, and scale normalization; Grayscale each of the first key frames to obtain a grayscale key frame; Convert each of the grayscale key frames into a grayscale image based on a programming language; Performing threshold segmentation on the grayscale image to obtain a binary image; Performing contour detection on the binary image based on an edge detection algorithm; Shape feature extraction is performed on the contour to obtain shape feature data, wherein the shape feature data includes but is not limited to contour features of Taekwondo action postures.
7. The method for detecting and analyzing Taekwondo movements and postures based on visual recognition as claimed in claim 1, characterized in that: The steps before building the action posture detection model include: Acquire a multi-feature fusion feature vector data set, and divide the multi-feature fusion feature vector data set into a multi-feature fusion feature vector training data set and a multi-feature fusion feature vector test data set; The motion posture detection model is trained based on the multi-feature fusion feature vector training data set, and the motion posture detection model is tested based on the multi-feature fusion feature vector test data set until the prediction accuracy of the motion posture detection model meets a preset accuracy threshold, and the training of the motion posture detection model is stopped.
8. A Taekwondo action posture detection and analysis system based on visual recognition, characterized in that: The system comprises: A video data module, used for collecting video data of the Taekwondo movements based on a camera module; A preprocessing module, used for preprocessing the video data to obtain first video data; A key frame extraction module, used for segmenting the first video data and extracting all key frames of the Taekwondo action; A feature extraction module is used to extract key point features from each key frame to obtain position feature data, angle feature data, motion trajectory feature data and shape feature data; A multi-feature fusion feature vector module is used to fuse the position feature data, the angle feature data, the motion trajectory feature data and the shape feature data to obtain a multi-feature fusion feature vector; The detection result module is used to construct an action posture detection model, input the multi-feature fusion feature vector into the action posture recognition model to detect Taekwondo action postures, and obtain Taekwondo action posture detection results.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the Taekwondo action posture detection and analysis method based on visual recognition described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the Taekwondo action posture detection and analysis method based on visual recognition according to any one of claims 1 to 7 are implemented.
