Sled starting stage motion analysis method and device

Abstract

The invention provides a motion analysis method and device for a sleigh starting stage. The method comprises the following steps: acquiring a video sequence of a to-be-analyzed target in the sleigh starting stage in a sleigh motion process; identifying a to-be-analyzed target in each video frame in the video sequence, and determining motion track information of the to-be-analyzed target; performing human body key point detection on the to-be-analyzed target in each video frame to obtain human body key point information of the to-be-analyzed target in each video frame; and determining a motion analysis result of the to-be-analyzed target based on the motion track information, the human body key point information and the moment information. The human body key point information of the to-be-analyzed target is determined based on the human body key point detection of the to-be-analyzed target, and the motion analysis of the to-be-analyzed target is performed based on the obtained data, so that the automatic motion analysis process of the to-be-analyzed target is realized, and the accurate recognition and tracking of the motion of an athlete can be realized. The accuracy of the analysis result is improved, and the subjectivity of manual judgment is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of motion analysis, and in particular to a motion analysis method and device for a sled start phase. Background Art

[0002] In order to scientifically collect athlete parameters and conduct motion analysis in bobsleigh analysis, existing methods are generally based on motion capture technology, including infrared motion capture technology, inertial wearable capture technology, and video analysis technology.

[0003] Existing motion capture technology has expensive equipment and complicated equipment operation procedures. Wearable devices will affect the training of luge athletes and cause signal interference, no video footage, and no displacement indicators, resulting in low accuracy in athlete motion analysis. Summary of the invention

[0004] The present invention provides a motion analysis method and device for a sled start phase, so as to improve the accuracy of the analysis of the sled motion process.

[0005] The present invention provides a motion analysis method for a sled start phase, comprising the following steps: Acquire a video sequence of the target to be analyzed during the sled departure phase of the sled movement; Identifying the target to be analyzed in each video frame in the video sequence, and determining the motion trajectory information of the target to be analyzed based on the target to be analyzed identified in each video frame; Performing human body key point detection on the target to be analyzed in each video frame to obtain human body key point information of the target to be analyzed in each video frame; Based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the time information of the target to be analyzed in the sled departure phase, motion analysis is performed on the target to be analyzed to determine the motion analysis result of the target to be analyzed.

[0006] According to a motion analysis method for a sled start phase provided by the present invention, obtaining a video sequence of a target to be analyzed during the sled start phase during the sled motion process includes: Based on multiple image acquisition devices and the time information of the target to be analyzed at the sled departure stage, synchronously acquire video data of multiple angles of the target to be analyzed at the sled departure stage; The video data from the multiple angles are fused to obtain fused video data, and video frames are extracted from the fused video to obtain the video sequence.

[0007] According to a motion analysis method for a sled departure phase provided by the present invention, based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the time information of the target to be analyzed in the sled departure phase, the motion analysis of the target to be analyzed is performed to determine the motion analysis result of the target to be analyzed, including: Determine the reaction speed of the target to be analyzed and the sled push speed of the target to be analyzed based on the motion trajectory information and the time information; Based on the human body key point information of the target to be analyzed in each video frame, determining the motion angle data of the target to be analyzed, wherein the motion angle data at least includes a joint angle, a trunk compression angle, and a body extension angle; Based on the reaction speed, the sled speed and the motion angle data, a motion analysis result of the target to be analyzed is determined.

[0008] According to a motion analysis method for a sled start phase provided by the present invention, the identifying of the target to be analyzed in each video frame in the video sequence comprises: Based on the target detection algorithm YOLO, the target to be analyzed in each video frame in the video sequence is identified, and the target detection frame of the target to be analyzed in each video frame is determined.

[0009] According to a motion analysis method for a sled start phase provided by the present invention, the motion trajectory information of the target to be analyzed is determined based on the target to be analyzed identified in each video frame, including: Based on the target tracking algorithm Deep SORT, target tracking is performed on the target detection frame in each video frame to determine the trajectory information of the target to be analyzed.

[0010] According to a motion analysis method for a sled start phase provided by the present invention, the human body key point detection is performed on the target to be analyzed in each video frame to obtain the human body key point information of the target to be analyzed in each video frame, including: Based on the HRNet key point detection technology, human key point detection is performed on the target to be analyzed in each video frame to obtain the human key point information of the target to be analyzed in each video frame.

[0011] The present invention also provides a motion analysis device for a sled start phase, comprising the following modules: A video sequence acquisition module, used to acquire a video sequence of the target to be analyzed during the sled departure phase of the sled movement; A trajectory determination module, used to identify the target to be analyzed in each video frame in the video sequence, and determine the motion trajectory information of the target to be analyzed based on the target to be analyzed identified in each video frame; A key point determination module is used to perform human key point detection on the target to be analyzed in each video frame to obtain human key point information of the target to be analyzed in each video frame; The motion analysis module is used to perform motion analysis on the target to be analyzed based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the time information of the target to be analyzed at the sled departure stage, and determine the motion analysis result of the target to be analyzed.

[0012] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the motion analysis method for the start phase of a sled as described above is implemented.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the motion analysis method for the sled departure phase as described in any one of the above is implemented.

[0014] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the motion analysis method for the start phase of a sled as described above is implemented.

[0015] The motion analysis method and device for the sled start phase provided by the present invention obtains the video sequence of the sled start phase, identifies and tracks the target to be analyzed in the video frame based on computer vision technology, and determines the trajectory information of the target to be analyzed. Video sequence analysis can be accurate to each frame and each action detail, thereby achieving in-depth analysis of the athlete's movements. Based on the detection of human key points of the target to be analyzed, the human key point information of the target to be analyzed is determined, and motion analysis of the target to be detected is performed based on the obtained data, thereby realizing the automatic motion analysis process of the target to be analyzed, and accurate identification and tracking of the athlete's movements can be achieved. The accuracy of the analysis results is improved and the subjectivity of human judgment is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is a schematic flow chart of the motion analysis method for the sled departure phase provided by the present invention.

[0018] Figure 2 It is a schematic diagram of the analysis process provided by the present invention.

[0019] Figure 3 It is a schematic diagram of the time constraint flow provided by the present invention.

[0020] Figure 4 It is a schematic diagram of analysis result feedback provided by the present invention.

[0021] Figure 5 It is a schematic diagram of the setting of the image acquisition device provided by the present invention.

[0022] Figure 6 It is a schematic diagram of the target tracking process provided by the present invention.

[0023] Figure 7 It is a schematic diagram of the human body key point information extraction process provided by the present invention.

[0024] Figure 8 The present invention is a schematic diagram of the structure of a device for applying the motion analysis method for the sled departure phase provided by the present invention.

[0025] Fig. 9 It is a structural schematic diagram of the motion analysis device for the sled departure phase provided by the present invention.

[0026] Fig.10 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0028] It provides new possibilities for the analysis of bobsleigh movements and postures. The technologies used in the related methods mainly include human body contour extraction technology, machine learning-based motion recognition technology, and human body 3D reconstruction technology based on depth information. It is common in video analysis that the analysis point cannot be accurately located due to obstruction, which is a shortcoming in video analysis technology. By synthesizing the video panorama of the analysis point, using adjacent pictures to determine the analysis point in the same video, and performing image analysis and automatic tracking more clearly according to the motion trajectory of the key points, the 3D analysis technology is made more scientific. But the problem is that there is no way to perform real-time data analysis during training and competition on the field, and give timely feedback to the coach, which increases the difficulty of equipment application. Machine learning in related methods is difficult to accurately capture the complex nonlinear structure of the human body and the spatiotemporal relationship between continuous motion frames. In practical applications, the human body is simplified into a rigid body connected by a series of rotating joints, and then the action is recognized and captured. However, human body movement is more complex, and organs such as bones and muscles cannot be replaced by simple rigid body models. In addition, the machine learning algorithm has complicated preprocessing procedures for kinematic data and limited processing capabilities, making it difficult to analyze the bobsleigh technical movements in a timely and accurate manner.

[0029] In order to scientifically collect the parameters of athletes under practical conditions, motion capture technology is widely used in bobsleigh analysis. The motion capture technologies in related methods mainly include infrared motion capture technology, inertial wearable capture technology and video analysis technology. Among them, infrared motion capture technology equipment is expensive, the equipment operation process is relatively complicated, and it is affected by light, so it is not applicable in real-time monitoring of bobsleigh. Inertial wearable capture technology uses wearable devices, which are generally troublesome to wear, and wearable devices will affect the training of bobsleigh athletes, and will also cause signal interference, no video footage, no displacement indicators, and low accuracy. The video analysis technology in the related method uses a high-speed camera to record the video, and then processes it through software, which requires a huge amount of manual work.

[0030] In addition, the biomechanical analysis of bobsleigh sports is very difficult due to its special characteristics such as fast instantaneous speed, high degree of track dispersion, long race distance, and low temperature non-laboratory venues. In view of its special sports form and sports environment, the wind tunnel flow field simulation test and computational simulation in the relevant methods provide a theoretical basis for technical characteristics of training. For example, in the wind tunnel simulation test, the biomechanical parameters of the athlete's body posture during the start-up acceleration phase were measured, and it was found that the incoordination of the ice-scraping action would affect the smoothness of the sliding, thereby helping the athlete to correct the wrong technical actions. Based on the wind tunnel test, Momose et al. conducted wind resistance wind tunnel tests on four bobsleigh sports postures, including lying flat, raising the head, tilting the head to the left / right, and raising the foot; CFD simulation was performed on the aerodynamic characteristics of the human / bobsleigh multi-body system during the sliding phase to explore the theoretical optimal ice-scraping sliding posture. When the bobsleigh project needs to raise the head to control and adjust the route during sliding, the head can be consciously tilted to reduce the adverse effects of wind resistance. The competition in the bobsleigh project is very fierce. The competition is timed at 0.001s. A 10% reduction in wind resistance during the sliding phase can improve the performance by 1%. Pointing toes and curling toes will lead to a 27% difference in CDA, which has a great impact on performance. Athletes should try to keep their toes tight when sliding, but the above theoretical research is relatively idealistic. In order to further explore real sports scenes, related methods also apply video panoramic analysis technology and ultra-wide band (UWB)-based biomechanical systems to scientific measurements of sleds, but video panoramic analysis has a large amount of manual labeling tasks and slow processing speed, making it difficult to provide real-time feedback for daily training; the UWB-based biomechanical system requires athletes to wear sensors throughout the entire exercise, which still has some impact on the aerodynamic characteristics of the human / sled multi-body system at each stage.

[0031] In view of the defects of the related methods, the present invention provides a motion analysis method for the sled departure stage. Figure 1 FIG. 1 is a flow chart of the motion analysis method for the sled departure stage provided by the present invention. Figure 1 As shown, the method includes the following: Step 110, obtaining a video sequence of the target to be analyzed during the sled departure phase of the sled movement; Step 120, identifying the target to be analyzed in each video frame in the video sequence, and determining the motion trajectory information of the target to be analyzed based on the target to be analyzed identified in each video frame; Step 130, performing human body key point detection on the target to be analyzed in each video frame to obtain human body key point information of the target to be analyzed in each video frame; Step 140, based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the time information of the target to be analyzed in the sled departure phase, perform motion analysis on the target to be analyzed to determine the motion analysis result of the target to be analyzed.

[0032] The execution subject of the motion analysis method for the sled departure phase provided by the present invention can be an electronic device, a component in an electronic device, an integrated circuit, or a chip. The electronic device can be a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA), etc., and the non-mobile electronic device can be a server, a network attached storage (NAS), or a personal computer (PC), etc., which is not specifically limited by the present invention.

[0033] The following takes the motion analysis method for the sled departure phase provided by the present invention executed by a computer as an example to describe the technical solution of the present invention in detail.

[0034] In step 110, a video sequence of the target to be analyzed during the sled departure phase of the sled movement is obtained.

[0035] Based on cameras at multiple angles, video data of the target at the start phase of the sledding process can be collected, and a video sequence can be constructed based on the collected video data.

[0036] Among them, the sled start phase during sledding can be divided into three parts, namely: 1. Push and pull the handle to push the sled into the track; 2. Use ice scraping gloves to paddle for about 7 meters; 3. Change to the best aerodynamic posture.

[0037] In step 120, the target to be analyzed in each video frame in the video sequence is identified, and based on the target to be analyzed identified in each video frame, the motion trajectory information of the target to be analyzed is determined.

[0038] Computer vision techniques (such as deep learning models) are used to automatically identify athletes in the video. This usually involves object detection algorithms such as YOLO (You Only Look Once) and SSD (Single Shot MultiBoxDetector), which can quickly and accurately locate the position of athletes in the video frames. By identifying the athletes in each frame, a continuous image sequence of the athletes throughout the start phase can be established.

[0039] Once the athlete is identified in each frame, the athlete's trajectory can be calculated through tracking algorithms. These algorithms are able to analyze the slight changes in the athlete's position between adjacent frames to construct the athlete's movement path throughout the start phase. The trajectory information includes key parameters such as speed, acceleration, displacement, etc., which is crucial for understanding the athlete's movement pattern.

[0040] In step 130, human key point detection is performed on the target to be analyzed in each video frame to obtain human key point information of the target to be analyzed in each video frame.

[0041] Human key point detection is another important technology in the field of computer vision. It can accurately mark the key parts of the human body (such as head, shoulders, elbows, knees, etc.) in the video frame. This is usually achieved through deep learning models that can identify and locate multiple key points of the human body. During the start phase of the sled, the information of the key points of the human body is crucial for evaluating the athlete's posture, balance and force distribution.

[0042] In step 140, based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the time information of the target to be analyzed in the sled departure phase, motion analysis is performed on the target to be analyzed to determine the motion analysis result of the target to be analyzed.

[0043] It should be noted that luge is a technical and tactical sport in which athletes lie on their backs on a sled and control their body's center of gravity to correct the sliding line on a specially designed ice track and complete a giant turn in the shortest possible time. Starting speed is one of the key factors affecting luge results, and starting time is also the most effective indicator for evaluating athletes' competitive performance at the starting stage.

[0044] The specific analysis process diagram can be shown as follows Figure 2 The analysis process provided by the present invention is shown in the schematic diagram. The motion trajectory information, human body key point information and time information (such as starting time, time when key actions occur, etc.) collected in the above steps are integrated for comprehensive analysis. This can include evaluating key actions such as the compression angle, straightening angle, ice scraping amplitude and ice scraping frequency when the athlete starts. By comparing the data of standard actions or excellent athletes, it can be further determined whether the sports performance of the target to be analyzed meets the standard and what potential room for improvement exists.

[0045] The sled start phase includes three parts, so the movement rate when pushing, pulling and scraping the ice, as well as the symmetry and stability of the bilateral limbs are important factors that determine the athlete's athletic performance. Focusing on the key link of the sled start, centering on the sled athlete's preparation and competition needs, the system analyzes the athlete's movement posture, speed, strength, sliding posture and other data at the start in real time, and presents them in a visual way to assist coaches and athletes in instantly judging the training status and movement accuracy, and correcting movement deviations. It can also summarize the technical performance and training progress of each athlete, providing a basis for coaches to formulate personalized training plans.

[0046] For the time information of the target to be analyzed during the bobsleigh start phase, a time constraint can be constructed. A time constraint is an accurate time measurement of an event within a specific time period. In the bobsleigh start phase, the time constraint is used to record the time from when the start signal is given to when the runner enters the track. The specific process can be as follows: Figure 3 The time constraint flow diagram provided by the present invention is shown in FIG.

[0047] Specifically include: High-precision timer: Hardware: Use high-precision timers, such as photoelectric gates, high-speed timers, etc. to record key time points. Trigger mechanism: Set up a trigger mechanism for the start signal to ensure that the timer starts timing when the signal is sent.

[0048] Video Sync: Timestamp: Add timestamp to each camera's video frame to ensure that the video stream is consistent with the time of the timer. Synchronization calibration: Ensure that the frame rate and timestamp of all cameras are synchronized through calibration to avoid time deviation.

[0049] Time Stamp: Key time points: Mark key time points in the video, such as the time when the starting signal is given, the time when the athlete starts pushing the sled, the time when the athlete enters the track, etc. Time difference calculation: Calculate the difference between each time point and generate a time report.

[0050] Data Analysis: Reaction time: The time difference between the start signal and the start of pushing the sled is calculated to evaluate the reaction speed. Pushing speed: The time difference between the start of pushing the sled and the entry into the track is calculated to evaluate the pushing speed.

[0051] Comprehensive Report: Generate detailed performance reports including reaction time, sled push speed, key point analysis, etc. to help coaches and athletes optimize training and competition strategies Based on the obtained motion analysis results, the training plan for the target to be analyzed can be supported. The specific process can be as follows: Figure 4 The analysis result feedback schematic diagram provided by the present invention is shown.

[0052] First, the video images are acquired through high-speed cameras. In order to avoid the impact of occlusion caused by the number and position of cameras on the capture of image detail features, and the impact of the limitation of calculation amount on the accuracy of motion capture, four high-speed cameras are used for fixed range shooting. Secondly, the data collected by the hardware equipment is sent to the software for analysis. The convolutional neural network deep learning model is mainly used to track the key points of the human body in the image. According to the key point data, the key action data such as compression angle, straightening angle, ice scraping amplitude and ice scraping frequency are calculated to support the training plan.

[0053] The motion analysis method for the sled start phase provided by the present invention obtains the video sequence of the sled start phase, identifies and tracks the target to be analyzed in the video frame based on computer vision technology, and determines the trajectory information of the target to be analyzed. Video sequence analysis can be accurate to each frame and each action detail, thereby achieving in-depth analysis of the athlete's movements. Based on the detection of human key points of the target to be analyzed, the human key point information of the target to be analyzed is determined, and motion analysis of the target to be detected is performed based on the obtained data, thereby realizing the automatic motion analysis process of the target to be analyzed, and accurate identification and tracking of the athlete's movements can be achieved. The accuracy of the analysis results is improved and the subjectivity of human judgment is reduced.

[0054] In one embodiment, a video sequence of a target to be analyzed during a sled departure phase during sled motion is obtained, including: synchronously obtaining video data of the target to be analyzed at multiple angles during the sled departure phase based on multiple image acquisition devices and time information of the target to be analyzed during the sled departure phase; fusing the video data at the multiple angles to obtain fused video data, and extracting video frames from the fused video to obtain the video sequence.

[0055] Optionally, four image acquisition devices may be provided, and the location diagram of the settings may be as follows: Figure 5 The image acquisition device provided by the present invention is shown in the schematic diagram. It is based on four image acquisition devices (cameras C1-C4) arranged around the starting stage of the sled, where G is the starting point of the target to be analyzed.

[0056] Based on the time information of the target to be analyzed during the sled departure phase, an accurate timestamp is added to the video data recorded by each camera to ensure that the video data from different angles can be accurately aligned to the same time point later.

[0057] Based on computer vision technology, the video frames from different cameras are registered, that is, the position, rotation and scaling between images are adjusted so that they are aligned in the same spatial coordinate system. Based on the results of image registration, the video data from multiple angles are fused and spliced ​​into one video to obtain fused video data.

[0058] From the fused video data, video frames are extracted according to a preset frame rate or key events (such as specific movements of athletes). The extracted video frames are arranged in chronological order to form a continuous video sequence. This video sequence will be used to analyze the details of the athletes' movements during the start phase of the sled.

[0059] In a real-time example, based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the moment information of the target to be analyzed in the sled departure phase, motion analysis is performed on the target to be analyzed to determine the motion analysis result of the target to be analyzed, including: determining the reaction speed of the target to be analyzed and the sled pushing speed of the target to be analyzed based on the motion trajectory information and the moment information; determining the motion angle data of the target to be analyzed based on the human body key point information of the target to be analyzed in each video frame, the motion angle data at least including the joint angle, the trunk compression angle, and the body extension angle; determining the motion analysis result of the target to be analyzed based on the reaction speed, the sled pushing speed, and the motion angle data.

[0060] Specifically, by analyzing the time difference between when an athlete receives the starting signal and when they start pushing the sled, the athlete's reaction speed can be calculated. This data is crucial for evaluating the athlete's agility and concentration.

[0061] Using the speed data in the motion trajectory information, we can analyze the acceleration and final speed of the athlete while pushing the sled. In particular, we focus on the speed changes of the athlete in the first few meters before departure. Combining reaction speed, sled pushing speed and movement angle data can provide a comprehensive assessment of the athlete's starting phase, including the athlete's power output, movement efficiency, body posture adjustment and other aspects.

[0062] In one embodiment, identifying the target to be analyzed in each video frame in the video sequence includes: based on the target detection algorithm YOLO, identifying the target to be analyzed in each video frame in the video sequence, and determining the target detection frame of the target to be analyzed in each video frame.

[0063] Based on the target to be analyzed identified in each video frame, the motion trajectory information of the target to be analyzed is determined, including: based on the target tracking algorithm Deep SORT, the target detection frame in each video frame is tracked to determine the trajectory information of the target to be analyzed.

[0064] Object tracking refers to the process of continuously detecting and locating a specific object in a video sequence. The object tracking process can be as follows: Figure 6The target tracking process provided by the present invention is shown in the schematic diagram. In the sled start phase, target tracking is used to monitor the entire process from the start to the entry of the contestant into the track. Specifically, it can be divided into the following three steps.

[0065] Object detection: Common detection algorithms include YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector), Faster R-CNN, etc. Take YOLO as an example, it is a real-time object detection algorithm that can complete the detection and positioning of the object through a single forward propagation. YOLO divides the input image into multiple grids, each of which is responsible for predicting a certain number of bounding boxes and their corresponding category probabilities.

[0066] Target recognition: Use convolutional neural networks (CNN) to extract target features. Common networks include ResNet, VGG, etc. The extracted features are input into the classifier to identify the category of the target.

[0067] Target tracking: Commonly used tracking algorithms include Deep SORT, FairMOT, Kalman Filter, etc. Taking DeepSORT as an example, it combines deep learning and Kalman filter to achieve robust multi-target tracking by maintaining a target's state estimation (position, speed, etc.) and appearance features. In each frame, by calculating the similarity between the candidate detection and the existing target (such as IOU, appearance feature distance, etc.), the detection result is matched with the target and the target's state is updated.

[0068] Based on the target tracking algorithm Deep SORT, the target detection box in each video frame is tracked to determine the trajectory information of the target to be analyzed.

[0069] In one embodiment, human key point detection is performed on the target to be analyzed in each video frame to obtain human key point information of the target to be analyzed in each video frame, including: based on HRNet key point detection technology, human key point detection is performed on the target to be analyzed in each video frame to obtain human key point information of the target to be analyzed in each video frame.

[0070] The extraction process of human key point information can be as follows: Figure 7 The human body key point information extraction process provided by the present invention is shown in the schematic diagram.

[0071] HRNet (High-Resolution Network) is a deep learning model designed for high-precision human key point detection. Traditional key point detection methods usually reduce the resolution of feature maps in the early stages to reduce computation and memory consumption, but this results in a decrease in the accuracy of key point detection. HRNet maintains high-resolution feature maps throughout the network and gradually introduces multi-scale feature fusion, thereby improving the robustness and generalization ability of the model while maintaining high accuracy.

[0072] The core idea of ​​HRNet is to maintain high-resolution feature maps throughout the network and enhance feature representation through multi-scale feature fusion. Specifically, HRNet consists of the following main parts: Initial layer: The initial part of the network usually includes a convolution layer and a maximum pooling layer to extract preliminary low-level features. The output feature map of this part has a low resolution but contains rich low-level information.

[0073] High-resolution branches: The core part of HRNet is multiple high-resolution branches, which maintain high-resolution feature maps at different stages. Each high-resolution branch extracts features through multiple residual blocks, which can effectively learn complex feature representations.

[0074] Multi-scale feature fusion: In order to introduce multi-scale information, HRNet adds new low-resolution branches in each stage and combines feature maps of different resolutions through cross-stage feature fusion. Specifically, the feature maps of the high-resolution branch are downsampled to the resolution of the low-resolution branch, while the feature maps of the low-resolution branch are upsampled to the resolution of the high-resolution branch. In this way, the high-resolution branch can utilize the contextual information of the low-resolution branch, while the low-resolution branch can benefit from the detail information of the high-resolution branch.

[0075] Final layer: At the final stage of the network, all feature maps of different resolutions are fused into a high-resolution feature map. This high-resolution feature map passes through a series of convolutional layers and upsampling operations to ultimately generate a heat map where each pixel value represents the probability of that location being a keypoint.

[0076] Based on the HRNet key point detection technology, human key point detection is performed on the target to be analyzed in each video frame to obtain the human key point information of the target to be analyzed in each video frame.

[0077] The present invention also provides a device using the motion analysis method for the sled departure phase provided by the present invention.

[0078] The device is mainly composed of three parts: one is the video data acquisition end, which is mainly composed of high-definition acquisition equipment (four monocular cameras); the second is the three-dimensional data analysis software platform end; and the third is the display device end, which is mainly composed of touch-screen all-in-one machines in the coach's seat and the athlete's channel.

[0079] The specific implementation process can be as follows Figure 8 The schematic diagram of the device structure using the motion analysis method for the sled departure phase provided by the present invention is shown.

[0080] First, we use 3D video motion capture and video feedback technology to collect video images of the technical movements of the bobsled athletes at the start stage. The collection process includes the collection of video images and the construction of a human motion posture recognition model, and the application of machine learning image feature recognition algorithms. The key parameter data is calculated and evaluated. The details are as follows: Video images were acquired through high-speed cameras. The number and placement of cameras should balance the occlusion caused by changes in viewing angles, the impact on the capture of image detail features, and the impact of computational limitations on the accuracy of motion capture, so as to adapt to the biomechanical characteristics and sports scenes of the acquisition targets. Four high-speed cameras were used to shoot the starting link, the moment of starting, and the posture characteristics of the athletes' ice-breaking movements. The angle range of the four main optical axes was 60° to 120°, so that the entire range of the athletes' activities was within the lens. After the position was fixed, the camera position was no longer adjusted until the shooting was completed. The frame rate was set to 60Hz, the shutter speed was 1 / 1000s, and the focus was manually adjusted. Before the formal shooting, a universal calibration frame was set up in the center of the athlete's activity range to ensure that the frame was fully presented in the four camera images, and the optimal occupancy ratio was 80% to 90%. The technical movements of the bobsleigh team athletes in the starting phase were shot by a high-definition camera system, and the trunk compression and body-lifting angle, upper limb explosive power analysis were performed through video analysis. The kinematic characteristics of the bobsleigh athletes' starting and ice-breaking technical movements: ice-breaking power, ice-breaking height, ice-breaking frequency, ice-breaking amplitude, arm length, etc. were analyzed. A high-definition camera is used to perform three-dimensional plane shooting at the starting position. The camera angle is perpendicular to the athlete's movement direction. The upper torso midline and horizontal compression angle and the upper torso midline and horizontal upright angle of the athlete at the start are analyzed.

[0081] 3D video analysis is the extraction of kinematic parameters of sports captured by 4 cameras. Through machine learning technology, various data of athletes collected are analyzed and the analysis results are displayed in a visual form, including charts, angle curves, starting speed, etc. In this way, the data of athletes' training conditions, sports performance, sports load, etc. in different cycles are counted, and the athletes' technical performance and training progress are summarized to provide a basis for coaches to formulate personalized training plans; finally, through the intelligent storage and analysis system, coaches are assisted in building a large database of athletes, and videos and data of athletes' training are stored in categories. By comparing the training performance of the same athlete at different times, the effectiveness of staged training can be effectively evaluated; by comparing the training performance of different athletes, it is convenient to see the gap and make targeted improvements. Track the sports performance of athletes in their growth stages.

[0082] The advantages of this device are: to avoid invasive equipment interfering with the training status of athletes, biomechanical analysis can meet the requirements of transparent and non-interference collection. Sports cameras or high-speed cameras shoot the movement process in a fixed area, and perform motion recognition on the captured images and videos. Motion recognition based on images and videos can obtain more comprehensive kinematic information. In addition to the sports silhouettes of athletes at different times, it can also obtain biomechanical parameters such as joint angles, movement posture characteristics, and speed of athletes at different times. Traditional manual labeling and machine learning data processing have limited computing power. The task of capturing human motion features in captured images is large and time-consuming, which to a certain extent limits the analysis and rapid feedback of technical features of movements. The deep learning adopted by this device has representation advantages and flexibility compared to traditional methods. Among various deep learning neural network models, its multi-task processing capability can perceive the dynamic characteristics of the spatial domain.

[0083] The motion analysis device for the sled start phase provided by the present invention is described below. The motion analysis device for the sled start phase described below and the motion analysis method for the sled start phase described above can correspond to each other.

[0084] like Fig. 9 As shown, the device comprises: The video sequence acquisition module 910 is used to acquire a video sequence of the target to be analyzed during the sled departure phase of the sled movement; A trajectory determination module 920 is used to identify the target to be analyzed in each video frame in the video sequence, and determine the motion trajectory information of the target to be analyzed based on the target to be analyzed identified in each video frame; A key point determination module 930 is used to perform human key point detection on the target to be analyzed in each video frame to obtain human key point information of the target to be analyzed in each video frame; The motion analysis module 940 is used to perform motion analysis on the target to be analyzed based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the time information of the target to be analyzed in the sled departure phase, and determine the motion analysis result of the target to be analyzed.

[0085] The motion analysis device for the sled start phase provided by the present invention obtains the video sequence of the sled start phase, identifies and tracks the target to be analyzed in the video frame based on computer vision technology, and determines the trajectory information of the target to be analyzed. The video sequence analysis can be accurate to each frame and each action detail, thereby achieving an in-depth analysis of the athlete's movements. Based on the detection of human key points of the target to be analyzed, the human key point information of the target to be analyzed is determined, and motion analysis of the target to be detected is performed based on the obtained data, thereby realizing an automatic motion analysis process of the target to be analyzed, and accurate identification and tracking of the athlete's movements can be achieved. The accuracy of the analysis results is improved and the subjectivity of human judgment is reduced.

[0086] In one embodiment, the video sequence acquisition module 910 is specifically used for: Obtain a video sequence of the target to be analyzed during the sled departure phase of the sled movement, including: Based on multiple image acquisition devices and the time information of the target to be analyzed at the sled departure stage, synchronously acquire video data of multiple angles of the target to be analyzed at the sled departure stage; The video data from the multiple angles are fused to obtain fused video data, and video frames are extracted from the fused video to obtain the video sequence.

[0087] In one embodiment, the motion analysis module 940 is specifically used to: Based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the time information of the target to be analyzed at the sled departure stage, the motion analysis of the target to be analyzed is performed to determine the motion analysis result of the target to be analyzed, including: Determine the reaction speed of the target to be analyzed and the sled push speed of the target to be analyzed based on the motion trajectory information and the time information; Based on the human body key point information of the target to be analyzed in each video frame, determining the motion angle data of the target to be analyzed, wherein the motion angle data at least includes a joint angle, a trunk compression angle, and a body extension angle; Based on the reaction speed, the sled speed and the motion angle data, a motion analysis result of the target to be analyzed is determined.

[0088] In one embodiment, the trajectory determination module 920 is specifically used to: Identifying the target to be analyzed in each video frame in the video sequence includes: Based on the target detection algorithm YOLO, the target to be analyzed in each video frame in the video sequence is identified, and the target detection frame of the target to be analyzed in each video frame is determined.

[0089] In one embodiment, the trajectory determination module 920 is further specifically configured to: Determining the motion trajectory information of the target to be analyzed based on the target to be analyzed identified in each video frame includes: Based on the target tracking algorithm Deep SORT, target tracking is performed on the target detection frame in each video frame to determine the trajectory information of the target to be analyzed.

[0090] In one embodiment, the key point determination module 930 is specifically used to: Performing human key point detection on the target to be analyzed in each video frame to obtain human key point information of the target to be analyzed in each video frame includes: Based on the HRNet key point detection technology, human key point detection is performed on the target to be analyzed in each video frame to obtain the human key point information of the target to be analyzed in each video frame.

[0091] Fig.10 An example of a physical structure diagram of an electronic device is shown in FIG. Fig.10 As shown, the electronic device may include: a processor 1010, a communication interface 1020, a memory 1030 and a communication bus 1040, wherein the processor 1010, the communication interface 1020 and the memory 1030 communicate with each other via the communication bus 1040. The processor 1010 may call the logic instructions in the memory 1030 to execute the motion analysis method of the sled departure stage, the method comprising: obtaining a video sequence of the target to be analyzed during the sled departure stage during the sled movement; Identifying the target to be analyzed in each video frame in the video sequence, and determining the motion trajectory information of the target to be analyzed based on the identified target to be analyzed in each video frame; Performing human body key point detection on the target to be analyzed in each video frame to obtain human body key point information of the target to be analyzed in each video frame; Based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the time information of the target to be analyzed in the sled departure phase, motion analysis is performed on the target to be analyzed to determine the motion analysis result of the target to be analyzed.

[0092] In addition, the logic instructions in the above-mentioned memory 1030 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0093] On the other hand, the present invention further provides a computer program product, the computer program product includes a computer program, the computer program can be stored in a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the motion analysis method of the sled departure phase provided by the above methods, the method comprising: obtaining a video sequence of the sled departure phase of the target to be analyzed during the sled motion process; Identifying the target to be analyzed in each video frame in the video sequence, and determining the motion trajectory information of the target to be analyzed based on the target to be analyzed identified in each video frame; Performing human body key point detection on the target to be analyzed in each video frame to obtain human body key point information of the target to be analyzed in each video frame; Based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the time information of the target to be analyzed in the sled departure phase, motion analysis is performed on the target to be analyzed to determine the motion analysis result of the target to be analyzed.

[0094] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to perform the motion analysis method of the sled departure phase provided by the above methods, the method comprising: obtaining a video sequence of the sled departure phase of the target to be analyzed during the sled motion process; Identifying the target to be analyzed in each video frame in the video sequence, and determining the motion trajectory information of the target to be analyzed based on the target to be analyzed identified in each video frame; Performing human body key point detection on the target to be analyzed in each video frame to obtain human body key point information of the target to be analyzed in each video frame; Based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the time information of the target to be analyzed in the sled departure phase, motion analysis is performed on the target to be analyzed to determine the motion analysis result of the target to be analyzed.

[0095] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0096] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A motion analysis method for a sled start phase, characterized in that: include: Acquire a video sequence of the target to be analyzed during the sled departure phase of the sled movement; Identifying the target to be analyzed in each video frame in the video sequence, and determining the motion trajectory information of the target to be analyzed based on the identified target to be analyzed in each video frame; Performing human body key point detection on the target to be analyzed in each video frame to obtain human body key point information of the target to be analyzed in each video frame; Based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the time information of the target to be analyzed in the sled departure phase, motion analysis is performed on the target to be analyzed to determine the motion analysis result of the target to be analyzed.

2. The motion analysis method for the sled start phase according to claim 1, characterized in that: The step of obtaining a video sequence of the target to be analyzed during the sledding start phase of the sledding movement includes: Based on multiple image acquisition devices and the time information of the target to be analyzed at the sled departure stage, synchronously acquire video data of multiple angles of the target to be analyzed at the sled departure stage; The video data from the multiple angles are fused to obtain fused video data, and video frames are extracted from the fused video to obtain the video sequence.

3. The motion analysis method for the sled start phase according to claim 1, characterized in that: The step of performing motion analysis on the target to be analyzed based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the time information of the target to be analyzed at the sled departure stage, and determining the motion analysis result of the target to be analyzed includes: Determine the reaction speed of the target to be analyzed and the sled push speed of the target to be analyzed based on the motion trajectory information and the time information; Based on the human body key point information of the target to be analyzed in each video frame, determining the motion angle data of the target to be analyzed, wherein the motion angle data at least includes a joint angle, a trunk compression angle, and a body extension angle; Based on the reaction speed, the sled speed and the motion angle data, a motion analysis result of the target to be analyzed is determined.

4. The motion analysis method for the sled start phase according to claim 1, characterized in that: The identifying the target to be analyzed in each video frame in the video sequence includes: Based on the target detection algorithm YOLO, the target to be analyzed in each video frame in the video sequence is identified, and the target detection frame of the target to be analyzed in each video frame is determined.

5. The motion analysis method for the sled start phase according to claim 4, characterized in that: The determining, based on the target to be analyzed identified in each video frame, the motion trajectory information of the target to be analyzed includes: Based on the target tracking algorithm Deep SORT, target tracking is performed on the target detection frame in each video frame to determine the trajectory information of the target to be analyzed.

6. The motion analysis method for the sled start phase according to claim 1, characterized in that: The detecting of human key points of the target to be analyzed in each video frame to obtain human key point information of the target to be analyzed in each video frame includes: Based on the HRNet key point detection technology, human key point detection is performed on the target to be analyzed in each video frame to obtain the human key point information of the target to be analyzed in each video frame.

7. A motion analysis device for a sled start phase, characterized in that: include: A video sequence acquisition module, used to acquire a video sequence of the target to be analyzed during the sled departure phase of the sled movement; A trajectory determination module, used to identify the target to be analyzed in each video frame in the video sequence, and determine the motion trajectory information of the target to be analyzed based on the target to be analyzed identified in each video frame; A key point determination module is used to perform human key point detection on the target to be analyzed in each video frame to obtain human key point information of the target to be analyzed in each video frame; The motion analysis module is used to perform motion analysis on the target to be analyzed based on the motion trajectory information, the human body key point information of the target to be analyzed in each video frame, and the time information of the target to be analyzed at the sled departure stage, and determine the motion analysis result of the target to be analyzed.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the motion analysis method for the sled start phase according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the motion analysis method for the sled start phase according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the motion analysis method for the sled start phase according to any one of claims 1 to 6 is implemented.

Images