Physical training posture acquisition method and system based on model database and adjustment method

Abstract

The invention discloses a physical training posture collection method and system based on a model database and an adjustment method, and belongs to the technical field of physical training, and the method comprises the following steps: S1, transmitting a single echo through a laser radar, and collecting human body point cloud data and image information in combination with a high-definition camera; s2, performing multi-dimensional maximum pooling operation on the point cloud data based on PointNet to generate a three-dimensional human body model; s3, extracting human body key points by using an RMPE technology, and combining the human body key points with the SMPL parameterized model; and S4, segmenting a multi-person view through Mask R-CNN, and carrying out comparative analysis on real-time action data and a champion model database. According to the physical training posture collection method and system based on the model database and the adjustment method, motion data are collected through a laser radar and a high-definition camera, a three-dimensional human body model is constructed by utilizing a point cloud processing technology, posture comparison and analysis are performed in combination with the champion model database, and low-delay feedback is realized through a real-time operation system.

Description

Technical Field

[0001] The present invention belongs to the technical field of sports training, and in particular relates to a sports training posture acquisition method, system and adjustment method based on a model database. Background Art

[0002] The intelligent sports training system is a digital training platform based on sensors, artificial intelligence and big data technologies. It collects sports posture, physiological indicators and training data in real time, combines AI algorithms to analyze movement standardization and physical condition, and provides personalized training plans and instant feedback. The system can simulate confrontation scenarios, assess sports injury risks, help athletes optimize technical details and adjust training intensity, and assist coaches in scientifically formulating periodized plans. It is widely used in professional competitions, campus sports and mass fitness, making the training process visual and precise, improving sports performance and reducing the probability of injury.

[0003] Existing sports training systems usually rely on single sensors or video analysis technology, making it difficult to accurately recognize the postures of multiple people in complex scenarios. They also lack real-time comparative analysis with standardized motion databases. Traditional methods have problems such as low data acquisition accuracy, high computational complexity, and significant feedback delays. They cannot meet the needs of real-time, efficient, and personalized sports training. Therefore, an integrated solution combining three-dimensional point cloud modeling, a champion model database, and real-time feedback is urgently needed. Summary of the Invention

[0004] In response to one or more of the above-mentioned defects or improvement needs in the prior art, the present invention provides a sports training posture acquisition method, system and adjustment method based on a model database, which has XXXX advantages.

[0005] To achieve the above object, the present invention provides a sports training posture acquisition method based on a model database, comprising the following steps: S1: Use laser radar to emit a single echo and combine it with a high-definition camera to collect human point cloud data and image information; S2: Perform multi-dimensional maximum pooling operations on point cloud data based on PointNet to generate a 3D human body model; S3: Extract human key points using RMPE technology and combine it with the SMPL parameterized model; S4: Segmenting multi-person views using Mask R-CNN and comparing real-time action data with the champion model database; S5: Output corrective suggestions based on the FreeRTOS real-time operating system.

[0006] Furthermore, the laser radar wavelength is 1550nm, and the amount of single echo point cloud data is not less than 90 million.

[0007] Furthermore, the champion model database contains motion parameters of several world champions and national champions, and body type, posture and shape are parameterized based on the SMPL model.

[0008] Furthermore, the real-time feedback system adjusts the detection parameters from small to large to eliminate false detections by setting a dynamic threshold condition function.

[0009] In a second aspect, the present invention further provides a sports training posture acquisition system based on a model database, comprising: Data acquisition module: including laser radar, high-definition camera and floor pressure sensor; Model processing module: configures the computing unit of PointNet and RMPE algorithms; Database module: cloud server that stores champion model parameters; Feedback module: interactive panel and display screen equipped with FreeRTOS.

[0010] Furthermore, the floor pressure sensor uses piezoelectric material to monitor footstep pressure and center of gravity change data in real time.

[0011] Furthermore, the interactive panel supports touch operation and integrates voice assistant and APP data synchronization functions.

[0012] In a third aspect, the present invention further provides a sports training posture adjustment method based on a model database, comprising the following steps: S1: Generate an exercise prescription report based on the deviation between real-time action and the champion model; S2: Perform machine learning analysis on long-term training data through the cloud platform to optimize personalized training plans; S3: Use smart wearable devices to monitor heart rate, cadence, and exercise intensity in real time, and dynamically adjust feedback content.

[0013] Furthermore, the exercise prescription report includes movement decomposition, posture assessment and health risk warning.

[0014] Furthermore, the smart wearable device has a built-in accelerometer, gyroscope and heart rate sensor, and the data is transmitted to the cloud platform through encryption.

[0015] In general, the above technical solutions conceived by the present invention have the following beneficial effects compared with the prior art: The sports training posture acquisition method, system and adjustment method based on the model database of the present invention collect motion data through laser radar and high-definition camera, use point cloud processing technology to build a three-dimensional human body model, combine with the champion model database to perform posture comparison analysis, and realize low-latency feedback through a real-time operating system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Flow chart of the embodiment of the present invention; Figure 2 This is a schematic diagram of PointNet point cloud data processing in the present invention; Figure 3 This is the principle diagram of RMPE attitude estimation of the present invention; Figure 4 This is a diagram of the hardware deployment of the present invention; Figure 5 Schematic diagram of the trajectory captured by the camera of the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0018] See also Figure 1-5 The sports training posture acquisition method based on the model database in this embodiment includes the following steps: S1: Use laser radar to emit a single echo and combine it with a high-definition camera to collect human point cloud data and image information; S2: Perform multi-dimensional maximum pooling operations on point cloud data based on PointNet to generate a 3D human body model; S3: Extract human key points using RMPE technology and combine it with the SMPL parameterized model; S4: Segmenting multi-person views using Mask R-CNN and comparing real-time action data with the champion model database; S5: Output corrective suggestions based on the FreeRTOS real-time operating system.

[0019] In this embodiment, the PointNet point cloud data processing method is a groundbreaking deep learning network in the prior art, designed specifically for processing 3D point cloud data. By performing maximum pooling operations on each dimension of the feature, PointNet can obtain the global features of the point cloud. In addition, PointNet's in-depth learning enables the system to achieve the optimal transformation matrix for point cloud alignment and solve the rotation and translation problems of the point cloud. Based on the PointNet point cloud data processing method, a maximum pooling operation in five dimensions is performed on the collected disordered point cloud to extract all point cloud features, and the chaotic point cloud is reasonably connected to generate a fixed-size feature vector for subsequent classification, detection or other tasks, providing a complete model for subsequent operations; Real-time Multi-Person Pose Estimation (RMPE) is a technology used to detect and analyze the poses of multiple people in images or videos in real time. It can identify and track key points on the bodies of multiple people, such as the head, wrists, shoulders, and knees. It is widely used in fields such as behavior recognition, health monitoring, augmented reality, and human-computer interaction. RMPE technology can quickly process image or video streams to achieve real-time pose estimation. Unlike methods that only estimate the pose of a single person, RMPE can simultaneously detect and estimate the poses of multiple people through images and videos captured by cameras. Based on RMPE pose estimation technology, the accuracy of motion recognition algorithms is improved and recorded in the system, enabling it to simultaneously standardize key points on multiple three-dimensional bodies and handle inaccurate human detection frames and redundant detections, thereby improving the accuracy of pose estimation and accurately identifying the movements and poses of primary and secondary school students in multi-person scenarios. Mask R-CNN is a deep learning architecture for target detection and instance segmentation. It can use feature pyramid networks to build high-resolution representations, which helps to detect targets of different sizes, and it can use RPN to quickly generate candidate target areas. Mask R-CNN can not only identify targets in images, but also perform pixel-level segmentation on each instance and output the target mask. In order to provide real-time feedback so that actions can be improved in a timely manner, the training and testing code is implemented based on the Mask R-CNN benchmark provided by Facebook AI Research. The Mask R-CNN learning architecture is combined with AI, enabling the system to use Mask R-CNN to detect and instance segment the different limbs of the user. The architecture enables the system to segment multiple human images in the same view, and decompose complex multi-person data into independent individual actions through the segmentation architecture, dividing it into many small calculations, reducing the amount of calculation for each action, greatly improving the calculation speed and recognition accuracy, based on Mask R-CNN. R-CNN performs pixel-level segmentation on the human body model constructed in a unified view, enabling separate calculations of each limb. This reduces data complexity, significantly reduces computational effort, and increases both computational speed and feedback speed. It uses motion analysis algorithms to accurately analyze each joint in the collected portraits, comparing and overlaying champion data from the champion model database with the movements of elementary and middle school students. It also incorporates OpenPose technology to implement a human pose analysis algorithm, which uses deep learning algorithms to detect and track key human points in images or videos, including those on the body, face, hands, and even feet. It relies on motion recognition algorithms to identify key human joints and accurately analyze the human body's poses. The motion recognition algorithm does not rely on specific detection instruments, but uses standard recognition devices as input devices to capture images or videos; and provides a calculation benchmark for OpenPose according to the joint points provided by the RMPE posture estimation technology, comparing and superimposing the detected human model movements with the parameters in the champion database, so that it can perform 3D posture analysis based on the 3D posture estimation method, solving the problem of difficulty in accurately identifying the movement trajectory of each limb of the user, accurately analyzing each user's movement process and movement data, and providing scientific guidance for physical education teaching.

[0020] Specifically, the laser radar wavelength is 1550nm, and the amount of single echo point cloud data is no less than 90 million.

[0021] In this embodiment, a carbon dioxide laser is used to emit a single echo based on high-precision human body scanning technology. Relying on the characteristics of the wave and the different reflection times of the received wave, 90 million distributed point cloud data are formed in the system. These data are exactly the contour curves of the human body, depicting the concave and convex parts of the human body in great detail. In order to minimize the harm of laser detectors to the eyes and other parts of primary and secondary school students, the wavelength of the continuous wave is adjusted to a uniform 1550nm to meet the standard of being harmless to primary and secondary school students.

[0022] Specifically, the champion model database contains the motion parameters of several world champions and national champions, and parameterizes body shape, posture and shape based on the SMPL model.

[0023] In this embodiment, in order to accurately compare the collected data with the data in the database, the SMPL model based on three-dimensional human body modeling and motion capture will use data parameters to represent the different shapes and postures of the human body. These parameters include body shape parameters, posture parameters, and shape parameters. In the construction of the champion model database, it can simulate the complex movements of the human body, including the protrusions and depressions of muscles during limb movement, avoiding surface distortion during movement. Based on the SMPL model, the optimal motion indicators of multiple world champions and national champions are used to construct a champion model database through parameters. The motion data in the champion model database has a more standard and secure motion reference, and the collected data can be quickly identified.

[0024] Specifically, the real-time feedback system adjusts the detection parameters from small to large to eliminate false detections by setting a dynamic threshold condition function.

[0025] In this embodiment, a real-time operating system (RTOS) is a specially designed operating system that ensures that tasks are completed within strict time limits. RTOSes are commonly used in control and data acquisition systems, which require high reliability and determinism. FreeRTOS supports multitasking, allowing multiple tasks (threads) to run concurrently. It can simultaneously calculate data for multiple limbs and multiple human bodies, achieving ideal low-volume, high-speed computing conditions. It also has low latency. The system can process tasks across multiple threads through its code, providing real-time feedback on the screen. By importing video into a database, a virtual environment is established, and the FreeRTOS real-time operating system code is introduced. The data obtained from the motion analysis algorithm is calculated and analyzed, and the results are displayed. When false detections cause errors in the video's human posture estimation, the code is set to increase the thresholds of the monitoring parameters in ascending order. This is repeated until false detections are eliminated, display errors disappear, and false feedback is reduced, achieving completely correct feedback. Ultimately, feedback is displayed on the screen as the display errors disappear. Conditional functions are then added to provide appropriate teaching guidance based on the completeness of the movement fit.

[0026] In a second aspect, the present invention further provides a sports training posture acquisition system based on a model database, comprising: Data acquisition module: including laser radar, high-definition camera and floor pressure sensor; Model processing module: configures the computing unit of PointNet and RMPE algorithms; Database module: cloud server that stores champion model parameters; Feedback module: interactive panel and display equipped with FreeRTOS In this embodiment, smart cameras installed around the classroom capture students' movements and postures in all directions. Equipped with advanced image recognition technology and deep learning algorithms, these cameras analyze students' movements in real time, providing accurate motion data. The smart cameras not only record students' movements but also process the data to generate three-dimensional motion models, providing detailed motion decomposition and posture assessment. The cameras integrate with other sensor systems within the classroom to enable simultaneous data collection and comprehensive analysis, improving the comprehensiveness and accuracy of monitoring.

[0027] Specifically, the floor pressure sensor uses piezoelectric materials to monitor footstep pressure and center of gravity change data in real time.

[0028] In this embodiment, pressure sensors placed on the classroom floor can accurately monitor footstep pressure and center of gravity changes. This data can be used to analyze the student's movement trajectory, gait, and force distribution, helping to optimize exercise posture and prevent sports injuries. The floor pressure sensors utilize technologies such as piezoelectric materials or resistance strain gauges, sensitive to pressure changes. This data is wirelessly transmitted in real time to a central data processing system for rapid analysis and feedback. Combined with a kinematics model, this system can provide students with a detailed gait analysis report, highlighting issues such as uneven gait and unstable center of gravity, helping them correct unhealthy exercise habits.

[0029] Specifically, the interactive panel supports touch operation and integrates voice assistant and APP data synchronization functions.

[0030] In this embodiment, the trainee's training status and data can be displayed in real time, which is convenient for the coach to provide on-site guidance and adjustments. The large-screen display system is seamlessly connected with the sensor network and cloud platform, and the display content is updated in real time. The display screen adopts high-definition resolution and large-size design to ensure that the data and images are clearly visible. The coach can quickly call up the trainee's training data through touch operation or voice control and provide detailed explanations and guidance.

[0031] In a third aspect, the present invention further provides a sports training posture adjustment method based on a model database, comprising the following steps: S1: Generate an exercise prescription report based on the deviation between real-time action and the champion model; S2: Perform machine learning analysis on long-term training data through the cloud platform to optimize personalized training plans; S3: Use smart wearable devices to monitor heart rate, cadence, and exercise intensity in real time, and dynamically adjust feedback content Specifically, the exercise prescription report includes movement breakdown, posture assessment and health risk warning.

[0032] In this embodiment, the above report can be fed back to parents, teachers and coaches through mobile applications. By viewing personalized training data and improvement suggestions on mobile phones, coaches can develop more targeted training plans based on the situation of each student. At the same time, the display screen in the classroom will also display the student's training status and data in real time, which is convenient for coaches to provide on-site guidance and adjustments. Through this multi-person data monitoring and cumulative evaluation, not only can comprehensive monitoring and analysis of students' personalized data be achieved, but also accurate and dynamic training management and effect evaluation can be provided, which greatly improves the training effect and student experience, making the training process more scientific, efficient and safe, and helping students better achieve their fitness goals.

[0033] Specifically, smart wearable devices have built-in accelerometers, gyroscopes and heart rate sensors, and the data is transmitted to the cloud platform through encryption.

[0034] In this embodiment, the smart wearable device has a built-in accelerometer, gyroscope and heart rate sensor, which can collect students' physiological indicators in real time, such as heart rate, blood oxygen saturation, body temperature and exercise data, such as number of steps, calories consumed and exercise intensity, to help students understand their exercise status and consumption level and provide data support; through intelligent algorithm analysis, personalized exercise suggestions can be provided, such as cadence adjustment, running posture optimization, etc., to improve training effects; the smart wearable device transmits data to the data processing center in real time via wireless connection; through the accumulation and analysis of students' long-term training data, the platform can generate personalized training reports and suggestions, forming long-term exercise records.

[0035] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A sports training posture acquisition method based on a model database, characterized in that: The following steps are involved: S1: Use laser radar to emit a single echo and combine it with a high-definition camera to collect human point cloud data and image information; S2: Perform multi-dimensional maximum pooling operations on point cloud data based on PointNet to generate a 3D human body model; S3: Extract human key points using RMPE technology and combine it with the SMPL parameterized model; S4: Segmenting multi-person views using Mask R-CNN and comparing real-time action data with the champion model database; S5: Output corrective suggestions based on the FreeRTOS real-time operating system.

2. The method according to claim 1, characterized in that The laser radar has a wavelength of 1550nm, and the amount of single echo point cloud data is no less than 90 million.

3. The method according to claim 1, characterized in that The champion model database contains the motion parameters of several world champions and national champions, and performs body type, posture and shape parameterization based on the SMPL model.

4. The method according to claim 1, wherein The real-time feedback system adjusts the detection parameters from small to large to eliminate false detections by setting a dynamic threshold condition function.

5. A sports training posture acquisition system based on a model database, characterized in that: include: Data acquisition module: including laser radar, high-definition camera and floor pressure sensor; Model processing module: configures the computing unit of PointNet and RMPE algorithms; Database module: cloud server that stores champion model parameters; Feedback module: interactive panel and display screen equipped with FreeRTOS.

6. The system according to claim 5, characterized in that The floor pressure sensor uses piezoelectric material to monitor footstep pressure and center of gravity change data in real time.

7. The system according to claim 5, characterized in that The interactive panel supports touch operation and integrates voice assistant and APP data synchronization functions.

8. A sports training posture adjustment method based on a model database, characterized in that: The following steps are involved: S1: Generate an exercise prescription report based on the deviation between real-time action and the champion model; S2: Perform machine learning analysis on long-term training data through the cloud platform to optimize personalized training plans; S3: Use smart wearable devices to monitor heart rate, cadence, and exercise intensity in real time, and dynamically adjust feedback content.

9. The method according to claim 8, characterized in that The exercise prescription report includes movement breakdown, posture assessment and health risk warning.

10. The method according to claim 8, characterized in that The smart wearable device has a built-in accelerometer, gyroscope and heart rate sensor, and the data is transmitted to the cloud platform through encryption.

Images