Track and field monitoring analysis system and analysis method based on machine vision

Abstract

The invention discloses a track and field sports monitoring analysis system and analysis method based on machine vision, and belongs to the field of athlete track and field sports intelligent analysis. The track and field sports monitoring analysis system comprises a multi-modal data acquisition module, a data preprocessing module, a feature extraction module, a model construction module and a comparison model library which are connected in sequence; the model construction module and the comparison model library are connected with the display module through the comparison analysis module, and a feedback generator is carried on the display module. By adopting the track and field sports monitoring analysis system and analysis method based on machine vision, the accuracy and reliability of sports data analysis are improved, scientific and reasonable training guidance is provided for athletes, and the technical level and competitive performance can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent analysis of athlete track and field sports, and in particular to a machine vision-based track and field sports monitoring and analysis system and analysis method. Background Art

[0002] With the rapid development of computer technology, sports analysis systems have gradually achieved digitalization and automation. Specifically, high-speed camera technology can be used to capture the athlete's movement trajectory, while sensors can be used to collect information such as the athlete's power output and physiological data. After analysis through computer algorithms, some relatively accurate kinematic and dynamic parameters can be obtained, providing a more scientific basis for the athlete's training.

[0003] In recent years, artificial intelligence (AI) technology has been widely used in the field of track and field technical analysis. Intelligent systems can automatically identify and analyze athletes' movements. Through machine learning algorithms, they learn and analyze large amounts of data, not only to identify problems with athletes' technical movements, but also to predict athletes' potential performances, providing more accurate advice to coaches and athletes. Some intelligent track and field technical analysis systems can even combine virtual reality (VR) and augmented reality (AR) technologies to provide athletes with an immersive training experience.

[0004] However, despite the significant progress these technologies have brought, they still face many challenges and shortcomings in practical applications:

[0005] 1. Excessive data volume: With the popularization of sensors and cameras, the amount of data collected has increased exponentially, bringing tremendous pressure to data analysis. Large amounts of data not only increase storage and processing costs, but may also lead to inefficient analysis.

[0006] 2. Data quality issues: Data may be inaccurate or incomplete, especially in outdoor environments, where factors such as weather and light can affect the quality of the data and thus the reliability of the analysis results.

[0007] 3. Accuracy in complex scenarios: Although AI has made progress in motion recognition and data prediction, accuracy still needs to be improved in complex motion scenarios (such as multi-person interaction, changing environments) and situations with large individual differences. For example, in terms of automatic key frame extraction, existing technologies sometimes fail to accurately capture important action moments.

[0008] 4. High cost limits popularization: Smart technology equipment is expensive and may be unaffordable for schools, small sports organizations or individuals. This limits the popularity and widespread application of these advanced technologies. For example, the high prices of depth cameras and high-end graphics cards make many organizations reluctant to buy them. Summary of the invention

[0009] The purpose of the present invention is to provide a track and field sports monitoring and analysis system and analysis method based on machine vision to solve the above technical problems.

[0010] To achieve the above-mentioned object, the present invention provides a track and field sports monitoring and analysis system based on machine vision, comprising a multimodal data acquisition module, a data preprocessing module, a feature extraction module, a model construction module and a comparison model library connected in sequence, the model construction module and the comparison model library are connected to a display module via a comparison analysis module, and a feedback generator is mounted on the display module;

[0011] Among them, the multimodal data acquisition module is used to collect the physical parameters and movement parameters of the key points of the athletes;

[0012] The data preprocessing module is used to preprocess the collected physical parameters and motion parameters of key points of athletes;

[0013] The feature extraction module is used to extract the pre-processed physical parameters and motion parameters of the key points of the athletes;

[0014] The model building module is used to build the athlete action model based on the extracted key point data;

[0015] The comparison model library is used to store standard action models;

[0016] The comparison and analysis module is used to compare the constructed athlete movement model with the standard movement model and output a difference report to the display module;

[0017] The display module is used to display the difference report and utilize the feedback generator to generate training suggestions based on the difference report.

[0018] Preferably, the multimodal data acquisition module includes an optical heart rate sensor for monitoring the athlete's heart rate and multiple groups of motion monitoring components for detecting the athlete's motion state. The optical heart rate sensor is worn on the athlete, and the multiple groups of motion monitoring components are arranged in sequence along the track and field competition path. Each group of motion monitoring components includes an acceleration sensor, a pressure sensor, an angular velocity sensor, an inertial sensor, and multiple high-definition cameras arranged at multiple angles for capturing the athlete's overall movements and multiple depth cameras for capturing the athlete's key movements. The acceleration sensor is used to capture the acceleration changes when the athlete is running, the pressure sensor is used to detect the pressure of the athlete on the ground when jumping, the angular velocity sensor is used to capture the rotation speed of the athlete's key points, and the inertial sensor is used to capture the athlete's posture.

[0019] Preferably, the key actions include the aerial posture when starting and jumping, and the shooting action when throwing;

[0020] Key points include the torso joint, shoulder joint, hip joint, knee joint, ankle joint, and elbow joint.

[0021] Preferably, both the high-definition camera and the depth camera are equipped with computer vision algorithms, which are used to identify the athlete's body parts, movement postures and movement trajectories.

[0022] An analysis method for a track and field sports monitoring and analysis system based on machine vision comprises the following steps:

[0023] S1, multimodal data acquisition and preprocessing;

[0024] S11, data collection: the athlete moves along the track and field competition path, and the multimodal data collection module starts working. At this time, the optical heart rate sensor is used to monitor the athlete's heart rate in real time and collect body parameters. The acceleration sensor collects the acceleration change a when running, the pressure sensor detects the ground pressure F when jumping, the angular velocity sensor collects the key point rotation speed ω, the inertial sensor collects the athlete's posture data, and uses high-definition cameras and depth cameras arranged at multiple angles to shoot the athlete. Then, based on the target detection algorithm, the athlete's body parts are identified, the key point position coordinates (x, y) are determined, and the action posture and motion trajectory information are obtained based on the tracking algorithm;

[0025] S12, data preprocessing: cleaning and denoising the collected data, and using filtering algorithms to remove noise interference for sensor data; grayscale and normalization processing are performed on image data to unify the data format;

[0026] S2, feature extraction: extract the body parameters and motion parameters of key points from the preprocessed data; at the same time, based on the image sequences obtained by the high-definition camera and the depth camera, use the computer vision algorithm to calculate the position coordinate changes of the athletes at different times and obtain the displacement information;

[0027] S3, calculating motion parameters, the motion parameters including step frequency f, stride S, jump height h and linear velocity v parameters;

[0028] The calculation formula of step frequency f is as follows:

[0029]

[0030] In the formula, n represents the number of footsteps on the ground; t represents the unit time;

[0031] The calculation formula of stride length S is as follows:

[0032]

[0033] Where (x1, y1( and (x2, y2) represent the position coordinates of the athlete when the consecutive steps land;

[0034] The calculation formula of jumping height h is as follows:

[0035] h = d2 - d1;

[0036] Wherein, d1 represents the height of the key points of the athlete's body during the jump when the athlete takes off, and the depth camera obtains the height of the key points of the athlete's body during the jump; d2 represents the height of the key points of the athlete's body during the jump when the athlete reaches the highest point;

[0037] The linear velocity v is calculated as follows:

[0038]

[0039] Where L represents the displacement of the athlete

[0040] S4. Using deep learning algorithms, build athlete movement models based on the extracted key point body parameters and motion parameters;

[0041] S5. Compare the constructed athlete movement model with the standard movement model in the comparison model library, and generate a difference report and training suggestions.

[0042] Preferably, the specific steps of shooting the athlete using a high-definition camera and a depth camera arranged at multiple angles in step S11, identifying the athlete's body parts based on a target detection algorithm, determining the key point position coordinates (x, y), and acquiring the action posture and motion trajectory information based on a tracking algorithm are as follows:

[0043] The first step is to input the images captured by the high-definition camera and the depth camera into the YOLO network. The YOLO network divides the input image into multiple grids, each of which is responsible for detecting targets in a specific area.

[0044] The second step is to input the image data with the athlete's body parts and key points into the YOLO network for training, so that the YOLO network can learn the feature representation of different parts and key points; when a new image enters, the YOLO network will predict whether there is a target in each grid, and if so, it will output the target's category and bounding box coordinates;

[0045] The third step is to decode the predicted bounding box and remove the bounding boxes with too high overlap by non-maximum suppression method to obtain the final accurate body part detection box;

[0046] Step 4: Determine the position coordinates of the key points in the detection frame according to the relative position relationship of the key points in the human body;

[0047] Step 5: Input the position coordinates of the determined key points into the tracking algorithm, assign an ID to each detected target, and extract the appearance feature vector of each target through the appearance feature extraction network to distinguish different targets;

[0048] Step 6: Mahalanobis distance D in each frame image M (x,y) Calculate the positional similarity between the detected target in the current frame and the tracked target:

[0049]

[0050] In the formula, x M and M Respectively represent the position coordinate vectors of the detected target and the tracked target in the current frame;

[0051] Using cosine distance D C (x c ,y c ) Calculate the similarity of appearance features between the detected target in the current frame and the tracked target:

[0052]

[0053] In the formula, x c and c Represent the appearance feature vectors of the detected target and the tracked target in the current frame respectively;

[0054] Step 7: Use the Hungarian algorithm to match the detection results of the current frame with the tracked targets in the previous frame to determine the motion trajectory of each target.

[0055] Therefore, the present invention adopts the above-mentioned track and field sports monitoring and analysis system and analysis method based on machine vision, which has the following beneficial effects:

[0056] 1. Multimodal data collection: (1) Comprehensiveness: The system not only collects the athlete's physical parameters (such as heart rate), but also collects motion parameters through a variety of sensors (accelerometers, pressure sensors, angular velocity sensors, inertial sensors) and high-definition cameras and depth cameras. This multimodal data collection method ensures all-round monitoring of the athlete's movements; (2) High precision: The optical heart rate sensor accurately monitors the athlete's heart rate changes; the high-definition camera and depth camera capture the athlete's overall movements and key movements from multiple angles, ensuring the accuracy and completeness of the data;

[0057] 2. Advanced data processing and feature extraction: (1) Preprocessing technology: The data preprocessing module can eliminate noise and outliers to ensure the accuracy of subsequent analysis; (2) Feature extraction: The system can accurately extract the body parameters and motion parameters of the athlete's key points (such as torso connection points, shoulder joints, hip joints, etc.), and use computer vision algorithms to identify body parts, movement postures and movement trajectories to build high-quality movement models. At the same time, only key point data is collected and processed, reducing the amount of data processing and processing pressure;

[0058] 3. Intelligent model construction and comparative analysis: (1) Personalized model: Based on the extracted key point data, the system constructs a personalized athlete movement model to truly reflect the athlete's actual movement situation; (2) Standard comparison: By comparing with the standard movement model stored in the comparison model library, the system can generate a detailed difference report to help coaches and athletes find problems in the movement; (3) Real-time feedback: The display module not only displays the difference report, but also uses the feedback generator to automatically generate training suggestions based on the difference report to provide instant guidance;

[0059] 4. Scientific training suggestions: (1) Highly targeted: Based on the specific problems in the difference report, the system generates personalized training suggestions to help athletes improve their body posture when starting, their aerial posture when jumping, and their throwing movements when throwing; (2) Quantitative indicators: The system calculates the athlete's cadence, stride, jump height, angular velocity, linear velocity and other quantitative indicators, providing clear goals and basis for training;

[0060] 5. Technical support and wide application: (1) Computer vision algorithm: The computer vision algorithm installed on the high-definition camera and depth camera enhances the system's intelligent recognition capability, with advantages such as high capture accuracy and low cost; (2) Strong applicability: The system is not only suitable for motion monitoring on track and field competition paths, but can also be extended to other types of sports, and has broad application prospects.

[0061] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 The present invention is a structural block diagram of a track and field sports monitoring and analysis system based on machine vision. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical scheme and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention are further described in detail in combination with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not used to limit the embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions.

[0064] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or inherent to these processes, methods, products or devices.

[0065] The embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0066] like Figure 1 As shown, a machine vision-based track and field sports monitoring and analysis system includes a multimodal data acquisition module, a data preprocessing module, a feature extraction module, a model construction module and a comparison model library connected in sequence, the model construction module and the comparison model library are connected to the display module via the comparison analysis module, and the display module is equipped with a feedback generator; wherein the multimodal data acquisition module is used to collect the body parameters and motion parameters of the key points of the athletes; the data preprocessing module is used to preprocess (clean and denoise) the collected body parameters and motion parameters of the key points of the athletes; the feature extraction module is used to extract the preprocessed body parameters and motion parameters of the key points of the athletes; the model construction module is used to construct the athlete action model based on the extracted key point data; the comparison model library is used to store the standard action model; the comparison analysis module is used to compare the constructed athlete action model with the standard action model and output a difference report to the display module; the display module is used to display the difference report, and use the feedback generator to generate training suggestions based on the difference report.

[0067] The multimodal data acquisition module includes an optical heart rate sensor for monitoring the athlete's heart rate and multiple groups of motion monitoring components for detecting the athlete's motion state. The optical heart rate sensor is worn on the athlete. Multiple groups of motion monitoring components are arranged in sequence along the track and field competition path. Each group of motion monitoring components includes an acceleration sensor, a pressure sensor, an angular velocity sensor, an inertial sensor, and multiple high-definition cameras arranged at multiple angles for shooting the athlete's overall movements and multiple depth cameras for shooting the athlete's key movements. The acceleration sensor is used to collect the acceleration changes of the athlete when running, the pressure sensor is used to detect the pressure of the athlete on the ground when jumping, the angular velocity sensor is used to collect the rotation speed of the athlete's key points, and the inertial sensor is used to collect the athlete's posture. Key movements include starting, aerial posture during jumping, and hand movements during throwing; key points include trunk connection points, shoulder joints, hip joints, knee joints, ankle joints, and elbow joints.

[0068] Both the high-definition camera and the depth camera are equipped with computer vision algorithms, which are used to identify the athletes' body parts, movement postures and movement trajectories.

[0069] An analysis method for a track and field sports monitoring and analysis system based on machine vision comprises the following steps:

[0070] S1, multimodal data acquisition and preprocessing;

[0071] S11, data collection: the athlete moves along the track and field competition path, and the multimodal data collection module starts working. At this time, the optical heart rate sensor is used to monitor the athlete's heart rate in real time and collect body parameters. The acceleration sensor collects the acceleration change a when running, the pressure sensor detects the ground pressure F when jumping, the angular velocity sensor collects the key point rotation speed ω, the inertial sensor collects the athlete's posture data, and uses high-definition cameras and depth cameras arranged at multiple angles to shoot the athlete. Then, based on the target detection algorithm, the athlete's body parts are identified, the key point position coordinates (x, y) are determined, and the action posture and motion trajectory information are obtained based on the tracking algorithm;

[0072] The specific steps of using high-definition cameras and depth cameras arranged at multiple angles to shoot athletes in step S11, identifying athlete body parts based on target detection algorithms, determining key point position coordinates (x, y), and obtaining action posture and motion trajectory information based on tracking algorithms are as follows:

[0073] The first step is to input the images captured by the high-definition camera and the depth camera into the YOLO network. The YOLO network divides the input image into multiple grids, each of which is responsible for detecting targets in a specific area.

[0074] The second step is to input the image data with the athlete's body parts and key points into the YOLO network for training, so that the YOLO network can learn the feature representation of different parts and key points; when a new image enters, the YOLO network will predict whether there is a target in each grid, and if so, it will output the target's category and bounding box coordinates;

[0075] The third step is to decode the predicted bounding box and remove the bounding boxes with too high overlap by non-maximum suppression method to obtain the final accurate body part detection box;

[0076] Step 4: Determine the position coordinates of the key points in the detection frame according to the relative position relationship of the key points in the human body;

[0077] Step 5: Input the position coordinates of the determined key points into the tracking algorithm, assign an ID to each detected target, and extract the appearance feature vector of each target through the appearance feature extraction network to distinguish different targets;

[0078] Step 6: Mahalanobis distance D in each frame image M (x,y) Calculate the positional similarity between the detected target in the current frame and the tracked target:

[0079]

[0080] In the formula, x M and M Respectively represent the position coordinate vectors of the detected target and the tracked target in the current frame;

[0081] Using cosine distance D C (x c ,y c ) Calculate the similarity of appearance features between the detected target in the current frame and the tracked target:

[0082]

[0083] In the formula, x c and c Represent the appearance feature vectors of the detected target and the tracked target in the current frame respectively;

[0084] Step 7: Use the Hungarian algorithm to match the detection results of the current frame with the tracked targets in the previous frame to determine the motion trajectory of each target.

[0085] S12, data preprocessing: cleaning and denoising the collected data, and using filtering algorithms to remove noise interference for sensor data; grayscale and normalization processing are performed on image data to unify the data format;

[0086] S2, feature extraction: extract the body parameters and motion parameters of key points from the preprocessed data; at the same time, based on the image sequences obtained by the high-definition camera and the depth camera, use the computer vision algorithm to calculate the position coordinate changes of the athletes at different times and obtain the displacement information;

[0087] S3, calculating motion parameters, the motion parameters including step frequency f, stride S, jump height h and linear velocity v parameters;

[0088] The calculation formula of step frequency f is as follows:

[0089]

[0090] In the formula, n represents the number of footsteps on the ground; t represents the unit time;

[0091] The calculation formula of stride length S is as follows:

[0092]

[0093] Where (x1, y1) and (x2, y2) represent the position coordinates of the athlete when the consecutive steps land;

[0094] The calculation formula of jumping height h is as follows:

[0095] h = d2 - d1;

[0096] Wherein, d1 represents the height of the key points of the athlete's body during the jump when the athlete takes off, and the depth camera obtains the height of the key points of the athlete's body during the jump; d2 represents the height of the key points of the athlete's body during the jump when the athlete reaches the highest point;

[0097] The linear velocity v is calculated as follows:

[0098]

[0099] Where L represents the displacement of the athlete

[0100] S4. Using deep learning algorithms, build athlete movement models based on the extracted key point body parameters and motion parameters;

[0101] S5. Compare the constructed athlete movement model with the standard movement model in the comparison model library, and generate a difference report and training suggestions.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A track and field sports monitoring and analysis system based on machine vision, characterized in that: It includes a multimodal data acquisition module, a data preprocessing module, a feature extraction module, a model construction module and a comparison model library which are connected in sequence. The model construction module and the comparison model library are connected to the display module via the comparison analysis module. The display module is equipped with a feedback generator. Among them, the multimodal data acquisition module is used to collect the physical parameters and movement parameters of the key points of the athletes; The data preprocessing module is used to preprocess the collected physical parameters and motion parameters of key points of athletes; The feature extraction module is used to extract the pre-processed physical parameters and motion parameters of the key points of the athletes; The model building module is used to build the athlete action model based on the extracted key point data; The comparison model library is used to store standard action models; The comparison and analysis module is used to compare the constructed athlete movement model with the standard movement model and output a difference report to the display module; The display module is used to display the difference report and utilize the feedback generator to generate training suggestions based on the difference report.

2. A machine vision-based track and field sports monitoring and analysis system according to claim 1, characterized in that: The multimodal data acquisition module includes an optical heart rate sensor for monitoring the athlete's heart rate and multiple groups of motion monitoring components for detecting the athlete's motion status. The optical heart rate sensor is worn on the athlete, and the multiple groups of motion monitoring components are arranged in sequence along the track and field competition path. Each group of motion monitoring components includes an acceleration sensor, a pressure sensor, an angular velocity sensor, an inertial sensor, and multiple high-definition cameras arranged at multiple angles for capturing the athlete's overall movements and multiple depth cameras for capturing the athlete's key movements. The acceleration sensor is used to capture the acceleration changes of the athlete when running, the pressure sensor is used to detect the pressure of the athlete on the ground when jumping, the angular velocity sensor is used to capture the rotation speed of the athlete's key points, and the inertial sensor is used to capture the athlete's posture.

3. A machine vision-based track and field sports monitoring and analysis system according to claim 2, characterized in that: Key movements include starting, aerial posture when jumping, and release when throwing; Key points include the torso joint, shoulder joint, hip joint, knee joint, ankle joint, and elbow joint.

4. A machine vision-based track and field sports monitoring and analysis system according to claim 3, characterized in that: Both the high-definition camera and the depth camera are equipped with computer vision algorithms, which are used to identify the athletes' body parts, movement postures and movement trajectories.

5. The analysis method of the machine vision-based track and field sports monitoring and analysis system according to claim 4, characterized in that: The following steps are involved: S1, multimodal data acquisition and preprocessing; S11, data collection: the athlete moves along the track and field competition path, and the multimodal data collection module starts working. At this time, the optical heart rate sensor is used to monitor the athlete's heart rate in real time and collect body parameters. The acceleration sensor collects the acceleration change a when running, the pressure sensor detects the ground pressure F when jumping, the angular velocity sensor collects the key point rotation speed ω, the inertial sensor collects the athlete's posture data, and uses high-definition cameras and depth cameras arranged at multiple angles to shoot the athlete. Then, based on the target detection algorithm, the athlete's body parts are identified, the key point position coordinates (x, y) are determined, and the action posture and motion trajectory information are obtained based on the tracking algorithm; S12, data preprocessing: cleaning and denoising the collected data and using filtering algorithms to remove noise interference for sensor data; For image data, grayscale and normalization processing are performed to unify the data format; S2, feature extraction: extract the body parameters and motion parameters of key points from the preprocessed data; at the same time, based on the image sequences obtained by the high-definition camera and the depth camera, use the computer vision algorithm to calculate the position coordinate changes of the athletes at different times and obtain the displacement information; S3, calculating motion parameters, the motion parameters including step frequency f, stride S, jump height h and linear velocity v parameters; The calculation formula of step frequency f is as follows: In the formula, n represents the number of footsteps on the ground; t represents the unit time; The calculation formula of stride length S is as follows: Where (x1, y1) and (x2, y2) represent the position coordinates of the athlete when the consecutive steps land; The calculation formula of jumping height h is as follows: h = d2 - d1; Wherein, d1 represents the height of the key points of the athlete's body during the jump when the athlete takes off, and the depth camera obtains the height of the key points of the athlete's body during the jump; d2 represents the height of the key points of the athlete's body during the jump when the athlete reaches the highest point; The linear velocity v is calculated as follows: Where L represents the displacement of the athlete S4. Using deep learning algorithms, build athlete movement models based on the extracted key point body parameters and motion parameters; S5. Compare the constructed athlete movement model with the standard movement model in the comparison model library, and generate a difference report and training suggestions.

6. A machine vision-based track and field sports monitoring and analysis system according to claim 5, characterized in that: The specific steps of using high-definition cameras and depth cameras arranged at multiple angles to shoot athletes in step S11, identifying athlete body parts based on target detection algorithms, determining key point position coordinates (x, y), and obtaining action posture and motion trajectory information based on tracking algorithms are as follows: The first step is to input the images captured by the high-definition camera and the depth camera into the YOLO network. The YOLO network divides the input image into multiple grids, each of which is responsible for detecting targets in a specific area. The second step is to input the image data with the athlete's body parts and key points into the YOLO network for training, so that the YOLO network can learn the feature representation of different parts and key points; when a new image enters, the YOLO network will predict whether there is a target in each grid, and if so, it will output the target's category and bounding box coordinates; The third step is to decode the predicted bounding box and remove the bounding boxes with too high overlap by non-maximum suppression method to obtain the final accurate body part detection box; Step 4: Determine the position coordinates of the key points in the detection frame according to the relative position relationship of the key points in the human body; Step 5: Input the position coordinates of the determined key points into the tracking algorithm, assign an ID to each detected target, and extract the appearance feature vector of each target through the appearance feature extraction network to distinguish different targets; Step 6: Mahalanobis distance D in each frame image M (x,y) Calculate the positional similarity between the detected target in the current frame and the tracked target: In the formula, x M and M Respectively represent the position coordinate vectors of the detected target and the tracked target in the current frame; Using cosine distance D C (x c ,y c ) Calculate the similarity of appearance features between the detected target in the current frame and the tracked target: In the formula, x c and c Represent the appearance feature vectors of the detected target and the tracked target in the current frame respectively; Step 7: Use the Hungarian algorithm to match the detection results of the current frame with the tracked targets in the previous frame to determine the motion trajectory of each target.

Images