Obstacle running visual detection method, device and equipment and storage medium

Abstract

The invention provides an obstacle running visual detection method, device and equipment and a storage medium, and the method comprises the steps: selecting an obstacle running subject, deploying cameras at two ends of a runway field of the obstacle running subject and beside each obstacle of the obstacle field, and transmitting videos shot by the cameras to video analysis equipment; detecting in the video analysis equipment, distinguishing sports personnel from background personnel through a target recognition model, continuously tracking whether the sports personnel meet motion characteristics or not, extracting posture points of the correctly tracked sports personnel through a posture recognition model, marking the posture points in a video, and transmitting the video to the side end equipment for playing; the method comprises the following steps: determining motion time for tracking a correct motion person through a target identification model, calculating attitude data based on a sequence of attitude points, respectively comparing and analyzing the attitude data with a standard specification and historical data of the motion person, and proposing a training scheme based on a comparison and analysis result and the motion time.

Description

TECHNICAL FIELD

[0001] The present document relates to the technical field of computer vision recognition, and particularly relates to an obstacle race visual detection method and device, equipment and a storage medium. BACKGROUND

[0002] 400-meter obstacle race is an important evaluation project in military sports, which requires passing through 2 100-meter tracks and crossing 16 obstacles, and the movement time and action standardization are core indicators of the project.

[0003] In the related art, the statistical movement time usually depends on the manual stop watch method, which can accurately record the overall time of the movement personnel from the initial starting to the final finish line. However, it is difficult to accurately record the individual movement time of each obstacle due to the influence of factors such as the distance of the site and the layout of the obstacles. The video recording method through the camera needs manual analysis of the video, which is time-consuming and makes it difficult to quickly obtain the results. In the judgment of the movement standardization of the movement personnel, the distance between the front and back of the obstacle race site is far, and the movement speed of the personnel is relatively fast. A large number of non-standard actions are easily missed due to ineffective observation. In addition, the background personnel in the site will affect the movement tracking. In other evaluation projects or schemes, only the movement personnel is tracked, but a part of the safety protection personnel exists in the movement process, which affects the movement tracking.

[0004] In summary, the above analysis of the development status of the technical field shows that the existing technology lacks a detection area for dividing each obstacle, and a scheme for dividing different personnel roles in the tracking process, accurately calculating the stage movement time based on a target recognition model, and automatically analyzing the standardization based on a key posture point identified by a posture recognition model. SUMMARY The purpose of the present application is to provide an obstacle race visual detection method, device, equipment and storage medium, which aims to solve the above problems in the prior art.

[0005] According to a first aspect of the embodiment of the present application, an obstacle race visual detection method is provided, comprising: Selecting an obstacle race subject, deploying a camera at both ends of the track site and beside each obstacle of the obstacle race subject, and transmitting the video captured by the camera to a video analysis device; Detecting in the video analysis device, distinguishing the movement personnel and the background personnel through a target recognition model, continuously tracking whether the movement personnel meets the movement characteristics, extracting the posture points of the correctly tracked movement personnel through a posture recognition model, and marking the posture points in the video and transmitting them to an edge device for playing; The target recognition model is used to determine the exercise time of the correct exercise personnel, and the posture data is calculated based on the exercise time sequence of the posture points, and the posture data is compared and analyzed with the standard specification and the historical data of the exercise personnel respectively, and the training scheme is proposed based on the comparison and analysis result and the exercise time.

[0006] According to a second aspect of the embodiment of the present application, a hurdle race visual detection device is provided, comprising: A pre-deployment module is configured to select a hurdle race subject, and deploy cameras at both ends of a track field of the hurdle race subject and beside each obstacle of an obstacle field, and transmit videos captured by the cameras to a video analysis device; A detection module is configured to detect in the video analysis device, distinguish exercise personnel and background personnel through a target recognition model, continuously track whether the exercise personnel meet exercise characteristics, and extract posture points of correct exercise personnel through a posture recognition model, and transmit videos with the posture points marked to an edge device for playing. A result analysis module is configured to determine exercise time of the correct exercise personnel through the target recognition model, calculate posture data based on the exercise time sequence of the posture points, compare and analyze the posture data with the standard specification and the historical data of the exercise personnel respectively, and propose a training scheme based on the comparison and analysis result and the exercise time.

[0007] According to a third aspect of the embodiment of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the computer program is executed by the processor, the steps of the hurdle race visual detection method provided in the first aspect of the present application are implemented.

[0008] According to a fourth aspect of the embodiment of the present application, a computer readable storage medium is provided, and information transmission implementation programs are stored on the computer readable storage medium, and when the programs are executed by a processor, the steps of the hurdle race visual detection method provided in the first aspect of the present application are implemented.

[0009] The technical scheme provided by the embodiment of the present application has the following beneficial effects: cameras are deployed at both ends of a track field of a hurdle race subject and beside each obstacle of an obstacle field, and all-around visual observation is achieved; in the detection process, exercise personnel and background personnel are distinguished, and the current object is continuously tracked to ensure that it is exercise personnel, so as to avoid recognition errors caused by visual interference; videos with posture points marked are played in real time in an edge device, the guidance of on-site exercise behavior is more efficient, and safety hazards can be found in time for on-site intervention; posture data is calculated based on the exercise time sequence of the posture points, the posture data is compared and analyzed with the standard specification and the historical data of the exercise personnel respectively, and the generated training scheme is more scientific and suitable for the current exercise personnel.

[0010] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to make one or more embodiments of the present specification or the technical solutions in the prior art clearer, the drawings needed in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments described in the present specification, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0012] Figure 1 is a flowchart of the obstacle running visual detection method of the embodiment of the present application; Figure 2 is a pre-deployment schematic diagram of the embodiment of the present application; Figure 3 is a schematic diagram of the obstacle side camera picture of the embodiment of the present application; Figure 4 is a schematic diagram of the cloud server storage of the embodiment of the present application; Figure 5 is a schematic diagram of the visual detection technical framework of the embodiment of the present application; Figure 6 is a schematic diagram of the obstacle running visual detection device of the embodiment of the present application; Figure 7 is a schematic diagram of the electronic device of the embodiment of the present application. DETAILED DESCRIPTION

[0013] In order to make one or more embodiments of the present specification or the technical solutions in the prior art clearer, the drawings needed in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments described in the present specification, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0014] Method embodiment According to the embodiment of the present application, an obstacle running visual detection method is provided, Figure 1 is a flowchart of the obstacle running visual detection method of the embodiment of the present application, as Figure 1 shown, the obstacle running visual detection method according to the embodiment of the present application specifically includes: In step S110, the obstacle run event is selected, cameras are arranged at both ends of the track field and beside each obstacle of the obstacle field of the obstacle run event, and videos captured by the cameras are transmitted to the video analysis device, specifically including: The obstacle run event selected is 400-meter obstacle run, wherein the 400-meter obstacle run consists of a 100-meter track field and a 100-meter obstacle field to form a closed loop field, the standard obstacle field includes hurdles, trenches, low walls, high board jumping platforms, horizontal ladders, single-log bridges, high walls, and low-pile nets, and the movement process needs to pass through the 100-meter track field twice and cross 16 obstacles; One camera is arranged at each end of the track field, and one camera is arranged beside each obstacle facing the obstacle and close to the outer side area of the track field, and a detection area is divided in the camera captured picture to avoid repeated detection with adjacent cameras, the edge line where the moving person enters in the detection area is the starting line, and the edge line where the moving person leaves is the terminal line, and preferably, a section of area corresponding to the starting line is the entry detection area, and a section of area corresponding to the terminal line is the exit detection area, so as to improve the recognition effect; Figure 2 is a pre-deployed schematic diagram of an embodiment of the present application, as shown in Figure 2 , which shows the 400-meter obstacle run closed loop field and the camera arrangement, the moving person in the forward direction starts from the starting point close to the No. 10 camera, runs 100 meters of the track, then turns around the flag, and then crosses the hurdles, trenches, low walls, high board jumping platforms, horizontal ladders, single-log bridges, high walls, and low-pile nets in turn; the moving person in the return direction crosses the low-pile nets, climbs the high walls, turns around the single-log bridge column, climbs the horizontal ladders, climbs the high board jumping platforms, drills through the low wall holes, jumps down the trenches, crosses the hurdles in five steps, turns around the flag, and then arrives at the 100-meter run terminal, the terminal line and the starting line are the same; The No. 1 camera and the No. 10 camera can capture the front half and the rear half of the 100-meter track field respectively, and the No. 2 camera to the No. 9 camera correspond to the eight obstacles one by one; Figure 3 is a schematic diagram of the camera picture beside the obstacle of an embodiment of the present application, as shown in Figure 3 , which shows the picture captured beside a certain obstacle, and preferably, the angle is adjusted to clearly capture the obstacle scene without capturing the track field, considering the position relationship between the safety protection personnel and the moving person in the field, the camera is arranged facing the obstacle and close to the outer side area of the track field, which is helpful to improve the visual detection effect of the moving person.

[0015] In step S120, it is detected in the video analysis device that the moving personnel and the background personnel are distinguished by the target recognition model, whether the moving personnel meet the motion characteristics is continuously tracked, the posture points of the correctly tracked moving personnel are extracted by the posture recognition model, and the posture points are labeled in the video and transmitted to the edge device for playing. Specifically, it includes: The video analysis device is connected to the camera in the venue through a network switch, reads the camera video through an RTSP / RTMP protocol, loads the target recognition model to infer the video frame by frame, and outputs the detection box, category, coordinate and other information of the target.

[0016] A pre-trained YOLO target recognition model is used for visual recognition. In the training process, picture materials under different light, time and motion states are collected, and the target to be visually recognized is labeled as training data set. The labeled categories are pedestrians, crossing piles, trenches, low walls, high board jumping platforms, horizontal ladders, single-log bridges, high walls, low pile nets and turning sign flags. The YOLO target recognition model is used to distinguish the moving personnel and the background personnel. The background personnel includes irrelevant interference personnel and safety protection personnel. The personnel outside the closed loop venue are marked as irrelevant interference personnel, the personnel inside the closed loop venue and appearing in the closed loop venue before the start of the motion for more than a threshold value are marked as safety protection personnel, and the personnel entering the detection area are marked as moving personnel. Preferably, since part of the obstacles are convex obstacles, the YOLO target recognition model cannot determine the depth information, and it may not be able to determine whether the personnel are inside or outside the closed loop venue. The content captured by the camera can be converted to a plane for further judgment, for example, in the form of Figure 2 .

[0017] The first target tracking algorithm is used to continuously mark the irrelevant interference personnel and the safety protection personnel, and the irrelevant interference personnel are continuously filtered. The first target tracking algorithm is the original ByteTracker algorithm, that is, it continuously tracks on the basis of YOLO recognition to avoid repeated identification of each frame. The second target tracking algorithm is used to track whether the moving personnel meet the motion characteristics. The second target tracking algorithm is the ByteTracker algorithm spliced with the classifier after the Kalman filter. The Kalman filter captures feature data, and the classifier outputs whether the motion characteristics are met. The moving personnel are continuously tracked based on the motion characteristics and the second target tracking algorithm. In the original ByteTracker algorithm, the Kalman filter is responsible for acquiring and processing feature data, and the classifier can automatically output whether the tracked target is a real moving personnel or a misjudged interference according to the captured feature data. In the embodiment of the present application, the classifier selects a lightweight XGBoost, which is convenient for practical application.

[0018] The safety protection personnel is necessary in the obstacle race, the possibility of interfering with the sports personnel is greater than that of other sports projects, so it is necessary to distinguish different kinds of personnel and continuously mark, preferably, for the personnel who do not meet the above-mentioned categories, a pop-up window prompts the investigation and treatment.

[0019] Assuming that the sports personnel pass through a certain obstacle area, the sports characteristics are entering the detection area, crossing the obstacle in the detection area, and leaving the detection area, in the embodiment of the application, the cameras work in turn according to the positions of the sports personnel, for example, after the 8th camera correctly tracks, the 7th camera starts to track, but if at this time two personnel enter the detection area from the entry detection area, the target tracking algorithm can track them respectively, the ByteTracker model maintains different target IDs, and the personnel without sports characteristics can be excluded, so as to correctly track while continuously tracking.

[0020] The posture recognition model is used to identify the key points of the head, neck, limbs and trunk of the correctly tracked sports personnel as posture points, and mark the coordinates of the posture points, in the embodiment of the application, the MMPose model is selected as the posture recognition model, that is, the posture points are further identified in the tracking detection box.

[0021] The edge device is equivalent to a device that can play video in real time, which is convenient for timely discovery of safety hazards and on-site intervention.

[0022] In step S130, the movement time of the correctly tracked sports personnel is determined by the target recognition model, the posture data is calculated based on the movement time sequence of the posture points, the posture data is compared and analyzed with the standard specification and the historical data of the sports personnel respectively, and the training scheme is proposed based on the comparison and analysis result and the movement time, which specifically includes: The movement time judgment of the traditional visual recognition needs manual analysis, and the previous mode cannot judge the movement time of a single stage, especially the movement time of a single obstacle, in the embodiment of the application, the detection area has been divided, that is, the starting line and the ending line are determined, so the calculation process of the movement time is: When the detection box of the target recognition model outputs the movement personnel and the edge line of the detection area is in contact, the ankle posture point in the limb key point is located, the time when the ankle posture point enters the detection area is taken as the starting time, and the time when the ankle posture point leaves the detection area is taken as the ending time; Preferably, if there is no obstacle in front of or behind the obstacle, the 100-meter turning flag is taken as the line, and in other cases, adaptive adjustment is made for division, so as to ensure that each link can be connected.

[0023] The difference between the ending time and the starting time is obtained, the segmented time performance corresponding to each camera is obtained, the sum of the segmented time performance is calculated as the total time performance, and the segmented time performance and the total time performance are both taken as the result of the movement time, and the total time performance is calculated by using formula 1: Formula 1; wherein, T represents the total time performance of 400-meter hurdle race, n represents the total number of cameras, represents the finish line time frame of the i segment, represents the start line time frame of the i segment.

[0024] The posture recognition model has recognized the key points including the head, neck, limbs and torso of the person tracking correct movements as posture points, and further calculated the posture data including body angles, speeds and force points, specifically including: The angle is calculated by the vector dot product formula, for example, the thigh vector = hip joint - knee joint, the calf vector = ankle joint - knee joint, and the included angle is calculated using arccos; the force point is approximated by the end points of the two ankle key points; the speed is determined by the time sequence change of the center point of the torso.

[0025] The posture data is analyzed with the preset threshold of the standard specification, and the degree of mastering the movement essentials, muscle labor degree, muscle strength and joint activity degree of the athletes are analyzed by comparison, so as to determine whether it is standard; The historical data of the corresponding athlete posture data is parsed from the cloud server, and the change degree analysis is performed, that is, the change analysis of the degree of mastering the movement essentials, muscle labor degree, muscle strength and joint activity degree, and the training scheme with pertinence is generated based on the comparison analysis result and the movement time.

[0026] Based on the posture point labeled video, only the segmented video from the starting line to the finish line of the athletes recognized by the target recognition model is retained, that is, most of the empty shot pictures without athletes are eliminated, and the segmented videos retained by each camera are spliced into a full video; The full video, each segmented video and posture data are stored in the cloud server, and the video and posture data are jointly used as the movement data, which are bound to the corresponding athlete identity and examination information, so as to serve as the basis of subsequent historical data. After login verification, the video data of any obstacle within a specified time period can be quickly viewed; Preferably, the historical movement data extracted from the cloud server can be loaded in the edge device and visually compared with the current movement situation; the movement time and the result of the specification analysis can also be maintained in the cloud server. In this case, the training scheme is directly generated in the cloud server, and the training scheme is also transmitted to the edge device for real-time guidance; Figure 4is a schematic diagram of a cloud server stored by an embodiment of the present application, as Figure 4 shown, the data storage process corresponding to the preferred scheme is shown, the gesture points are marked in the video by the video analysis device and played in real time in the edge device, the motion related result data is uploaded to the cloud server to generate a training scheme, and the edge device interaction device is displayed.

[0027] The above technical solutions of the embodiments of the present application are illustrated by combining the following drawings.

[0028] Figure 5 is a schematic diagram of a visual detection technical framework of an embodiment of the present application, as Figure 5 shown, the complete technical architecture of the obstacle running visual detection is shown, and the main implementation process includes loading information, deploying a camera and dividing a detection area, personnel visual detection, detecting safety protection personnel, filtering irrelevant interfering personnel, tracking the current motion personnel, gesture recognition and marking, motion specification analysis, historical motion data comparison and outputting a training scheme, etc.

[0029] In summary, in view of the existing problems, the obstacle running visual detection method of the present application deploys a camera at both ends of the track site of the obstacle running purpose and beside each obstacle, realizes omnidirectional visual observation; in the detection process, the motion personnel, other irrelevant personnel and safety protection personnel are distinguished, and the safety protection personnel is allowed to appear in the closed loop site, the current object is continuously tracked to ensure that it is motion personnel, and recognition errors caused by visual interference are avoided; the video marked with gesture points is played in real time in the edge device, the guidance of the on-site motion behavior is more efficient, and safety hazards can be found in time for on-site intervention; when calculating the motion time, a set detection area is extracted between each obstacle, the time is calculated by the difference between the video frames corresponding to the detection area, the end line and the starting line, the total time result and the segmented time result can be accurately calculated at the same time, and the guidance and analysis are facilitated; the gesture data is calculated based on the motion time sequence of the gesture points, and the gesture data is compared and analyzed with the standard specification and the historical data of the motion personnel, so that the generated training scheme is more scientific and suitable for the current motion personnel; the motion data is maintained in the cloud server and bound with the corresponding motion personnel, which is convenient for timely finding, and the videos captured by each camera are stored after removing the redundancy, which effectively reduces the storage cost.

[0030] Device embodiment According to the embodiments of the present application, an obstacle running visual detection device is provided, Figure 6 is a schematic diagram of an obstacle running visual detection device of an embodiment of the present application, as Figure 6 shown, the obstacle running visual detection device according to the embodiments of the present application specifically includes: The pre-deployment module 60 is used to select the obstacle course and deploy cameras at both ends of the obstacle course and next to each obstacle. The video captured by the cameras is transmitted to the video analysis equipment, specifically for: The selected obstacle course is the 400-meter obstacle course, which consists of a closed loop consisting of a 100-meter track and a 100-meter obstacle course. A camera is deployed at each end of the runway, and another camera is deployed in the outer area facing the obstacle and close to the runway. A camera is deployed for each obstacle, and the detection area is divided in the camera's field of view.

[0031] Detection module 62 is used in video analysis equipment to detect moving people and background people through a target recognition model, continuously track whether the moving people meet the motion characteristics, extract the posture points of the correctly tracked moving people through a posture recognition model, and mark the posture points in the video for transmission to the edge device for playback. Specifically, it is used for: The YOLO target recognition model distinguishes between athletes and background personnel. Background personnel include irrelevant interference personnel and security personnel. Personnel outside the closed-loop area are marked as irrelevant interference personnel. Personnel inside the closed-loop area who have been in the closed-loop area for more than a threshold time before the start of the exercise are marked as security personnel. Personnel entering the detection area are marked as athletes. The first target tracking algorithm is used to continuously mark irrelevant interference personnel and security personnel, and to continuously filter out irrelevant interference personnel. The first target tracking algorithm is the ByteTracker model. The second target tracking algorithm is used to track whether the moving person meets the motion characteristics. The second target tracking algorithm is the ByteTracker model, which is a classifier concatenated after the Kalman filter. The Kalman filter captures feature data, and the classifier outputs whether the motion characteristics are met.

[0032] The posture recognition model identifies key points, including the head, neck, limbs, and torso of a person in correct motion, as posture points.

[0033] The results analysis module 64 is used to determine the movement time of the correctly tracked athlete through the target recognition model, calculate posture data based on the movement time series of posture points, compare and analyze the posture data with standard specifications and the athlete's historical data, and propose a training plan based on the comparison analysis results and movement time. Specifically, it is used for: When the target recognition model outputs the detection box of the moving person and it touches the edge of the detection area, the ankle posture point is located. The time when the ankle posture point enters the detection area is taken as the start time, and the time when it leaves the detection area is taken as the end time. The termination time is subtracted from the starting time to obtain a segmented time result corresponding to each camera, the sum of the segmented time results is calculated as a total time result, and the segmented time result and the total time result are both taken as a result of the sports time.

[0034] The posture data includes body angle, speed and force point.

[0035] The posture data is analyzed according to a preset threshold of a standard specification, historical data corresponding to the posture data of the sports person is parsed from a cloud server, and a change degree is analyzed. Based on the video of the posture points, only the segmented video of the sports person from the starting line to the termination line identified by the target recognition model is reserved, the segmented video reserved by each camera is spliced into a whole video, and the whole video, the segmented video and the posture data are stored in the cloud server.

[0036] In summary, in view of the existing problems, the visual detection device for obstacle run is deployed with cameras at both ends of the track field and beside each obstacle in the obstacle field to realize omnidirectional visual observation; in the detection process, the sports person, other irrelevant persons and safety protection personnel are distinguished, and the safety protection personnel are allowed to appear in the closed loop field, the current object is continuously tracked to ensure that it is the sports person, and identification errors caused by visual interference are avoided; the video marked with the posture points is played in the edge device in real time, the guidance of the on-site sports behavior is more efficient, and safety hazards can be found in time for on-site intervention; when the sports time is calculated, a detection area is extracted between each obstacle, the time is calculated based on the difference between the video frames corresponding to the finish line and the starting line in the detection area, the total time result and the segmented time result can be accurately calculated at the same time, and the overall guidance and analysis are facilitated; the posture data is calculated based on the sports time sequence of the posture points, the posture data is compared and analyzed with the standard specification and the historical data of the sports person, respectively, the generated training scheme is more scientific and suitable for the current sports person; the sports data is maintained in the cloud server and bound to the corresponding sports person, the sports data can be found in time, and the videos captured by each camera are stored after removing the redundancy, thereby effectively reducing the storage cost.

[0037] Electronic device embodiment Figure 7 is a schematic diagram of an electronic device of an embodiment of the present application. The electronic device 700 can include at least one processor 710 and a memory 720. The processor 710 can execute instructions stored in the memory 720. The processor 710 is communicatively connected to the memory 720 through a data bus. In addition to the memory 720, the processor 710 can also be communicatively connected to an input device 730, an output device 740 and a communication device 750 through the data bus.

[0038] The processor 710 can be any conventional processor, such as commercially available CPUs. The processor can also include a graphics processing unit (GPU), a field programmable gate array (FPGA), a system on chip (SOC), an application specific integrated circuit (ASIC), or combinations thereof.

[0039] The memory 720 can be implemented by any type of volatile or nonvolatile storage devices or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read only memory (EEPROM), erasable programmable read only memory (EPROM), programmable read only memory (PROM), read only memory (ROM), magnetic storage, flash memory, magnetic or optical disk.

[0040] In the embodiments of the present disclosure, the memory 720 stores executable instructions, and the processor 710 can read the executable instructions from the memory 720 and execute the instructions to implement all or part of the steps of the obstacle running visual detection method in any of the above example embodiments.

[0041] Computer readable storage medium embodiments In addition to the above method and device, the example embodiments of the present disclosure can also be a computer program product or a computer readable storage medium storing the computer program product, the computer program product including computer program instructions executable by a processor to implement all or part of the steps described in the obstacle running visual detection method in any of the above example embodiments.

[0042] The computer program product can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++, and conventional procedural programming languages, such as the "C" programming language or similar programming languages and scripting languages (e.g., Python). The program code can execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device or entirely on the remote computing device or server.

[0043] The computer readable storage medium can take the form of one or more combinations of any type of computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium can include, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a computer readable storage medium include, for example, a static random access memory (SRAM), an electrically erasable programmable read only memory (EEPROM), a programmable read only memory (PROM), a read only memory (ROM), a magnetic storage, a flash memory, a disk, or an optical disk, or any suitable combination of the foregoing.

[0044] Finally, it should be noted that the above-described embodiments are merely intended to illustrate the technical solutions of the present application, and are not intended to limit the present application; even though the present application has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the above-described embodiments, or equivalently replace some or all of the technical features thereof; and such modifications or replacements do not cause the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A hurdle race vision detection method, characterized by, The application relates to a method for analyzing and training a selected obstacle race subject. The method comprises the following steps: deploying cameras at both ends of a track field and beside each obstacle in an obstacle field of the selected obstacle race subject, and transmitting videos captured by the cameras to a video analysis device; In the video analysis device, a target recognition model is used to distinguish between moving personnel and background personnel, and the moving personnel are continuously tracked to determine whether the moving personnel meet the movement characteristics; a posture recognition model is used to extract posture points of the correctly tracked moving personnel, and the posture points are marked in the video and transmitted to an edge device for playing; A target recognition model is used to determine the movement time of the correctly tracked moving personnel, posture data is calculated based on the sequence of the posture points, the posture data is compared with standard specifications and historical data of the moving personnel, and a training scheme is proposed based on the comparison result and the movement time.

2. The method of claim 1, wherein, The method comprises the following steps: The selected obstacle race subject is 400-meter obstacle race, wherein the 400-meter obstacle race comprises a 100-meter track field and a 100-meter obstacle field to form a closed loop field; One camera is arranged at each end of the track field, and cameras are arranged in the outer side region opposite to the obstacles and close to the track field; one camera is arranged for each obstacle, and a detection region is divided in the camera shooting picture.

3. The method of claim 1, wherein, The method comprises the following steps: A YOLO target recognition model is used to distinguish between moving personnel and background personnel, wherein the background personnel include irrelevant interference personnel and safety protection personnel; personnel outside the closed loop field are marked as irrelevant interference personnel; personnel inside the closed loop field and appearing in the closed loop field before the start of the movement for more than a threshold value are marked as safety protection personnel; personnel entering the detection region are marked as moving personnel; A first target tracking algorithm is used to continuously mark the irrelevant interference personnel and the safety protection personnel, and the irrelevant interference personnel are continuously filtered, wherein the first target tracking algorithm is a ByteTracker algorithm; A second target tracking algorithm is used to track whether the moving personnel meet the movement characteristics, wherein the second target tracking algorithm is a ByteTracker algorithm after a Kalman filter and a classifier, feature data are captured through the Kalman filter, and whether the moving personnel meet the movement characteristics is output through the classifier.

4. The method of claim 1, wherein, The method comprises the following steps:

5. The method of claim 1, wherein, The posture recognition model is used to identify key points including the head, neck, limbs and torso of the correctly tracked moving personnel as posture points. The method comprises the following steps: When the detection frame of the target recognition model and the edge line of the detection region are in contact, an ankle posture point is positioned; the time when the ankle posture point enters the detection region is used as the starting time, and the time when the ankle posture point leaves the detection region is used as the ending time. The termination time is subtracted from the starting time to obtain a segment time result corresponding to each camera, the sum of the segment time results is calculated as a total time result, and the segment time result and the total time result are both taken as a result of the exercise time.

6. The method of claim 1, wherein, The calculating the posture data based on the sequence of the posture points specifically includes: calculating the posture data including body angles, speeds, and force points.

7. The method of claim 1, wherein, The comparing and analyzing the posture data with the standard specification and the historical data of the exercise personnel specifically includes: The posture data is analyzed according to a preset threshold of the standard specification, historical data of the posture data of the corresponding exercise personnel is parsed from a cloud server, and a change degree is analyzed. Based on the video marked by the posture points, only the segment video from the starting line to the termination line of the exercise personnel identified by the target recognition model is reserved, the segment videos reserved by each camera are spliced into a whole video, and the whole video, the segment videos, and the posture data are stored in the cloud server.

8. A hurdle race vision detection device, characterized by, The method comprises the following steps: The pre-deployment module is configured to select an obstacle race subject, deploy cameras at both ends of a track field of the obstacle race subject and beside each obstacle of an obstacle field, and transmit videos captured by the cameras to a video analysis device. The detection module is configured to detect in the video analysis device, distinguish exercise personnel and background personnel by a target recognition model, continuously track whether the exercise personnel meet exercise characteristics, extract posture points of the correctly tracked exercise personnel by a posture recognition model, mark the posture points in the video, and transmit the video to an edge device for playing. The result analysis module is configured to determine exercise time of the correctly tracked exercise personnel by the target recognition model, calculate posture data based on a sequence of the posture points, compare and analyze the posture data with a standard specification and historical data of the exercise personnel, and propose a training scheme based on a result of the comparison and analysis and the exercise time.

9. An electronic device, comprising: The computer program stored on the memory and executable on the processor implements the steps of the obstacle race visual detection method according to any one of claims 1 to 7 when executed by the processor. The computer readable storage medium stores an implementation program of information transmission, and the program implements the steps of the obstacle race visual detection method according to any one of claims 1 to 7 when executed by the processor.

10. A computer-readable storage medium, characterized in that, ​

Images