Campus intelligent playground system and motion detection method based on the system

Abstract

The present application relates to the field of smart playgrounds, and discloses a campus smart playground system and a motion detection method based on the system. The system includes an image acquisition device, a motion detection module, and a performance storage module arranged at each sports project area. The motion detection module uses the image acquisition device to collect sports videos for each sports project, tracks and detects multiple sports students in the sports video based on a multi-target tracking detection model to determine the students' multiple sports stages and key state frames of the corresponding stages, and segments the agreed human body parts and regular symbols in the key state frames based on the image segmentation model and generates the students' sports performance based on the segmentation results; the performance storage module stores the students' sports performance. The present application can not only automatically detect and identify the sports performance of different sports projects, but also simultaneously track and detect multiple sports students in the same sports project area and obtain the sports performance of multiple sports students at the same time.

Description

Technical Field

[0001] The present application relates to the field of smart playgrounds, and in particular to a campus smart playground system and a motion detection method. Background Art

[0002] In recent years, schools and parents have paid more and more attention to the all-round development of students. While ensuring that students have excellent academic performance, they also pay attention to students' physical fitness and health training. With the development of science and technology, more and more sports have realized intelligent detection, but the existing intelligent detection methods for sports events are only applicable to a single event and have low detection efficiency. For example, the Chinese invention patent with publication number CN113177476 A, "A method, system and test method for identifying key points of the heel of a standing long jump", first of all, this recognition scheme is only applicable to standing long jump and only supports single-person long jump detection. Faced with application scenarios with a large number of people in schools, the detection efficiency is low. In addition, there are currently many long jump detection methods based on posture detection, which involve many skeleton point parameters and have high computational complexity.

[0003] Furthermore, the drawback of testing methods that only apply to a single sport is that they typically operate independently, making data integration difficult. This often results in manual assessment, recording, and integration of each student's various sports. This method is not only time-consuming and labor-intensive, with inconvenient data recording, but also prevents intelligent utilization of the recorded data, making it impossible to conduct a comprehensive analysis of the student's various exercise and physical fitness data. Summary of the Invention

[0004] The purpose of this application is to provide a campus intelligent playground system and a motion detection method based on the system, which can not only automatically detect and identify the athletic performance of different types of sports, but also simultaneously track and detect multiple sports students in the same sports area and obtain the athletic performance of multiple sports students at the same time.

[0005] This application discloses a campus intelligent playground system, comprising:

[0006] An image acquisition device is provided at each sports area of ​​the playground;

[0007] A motion detection module is configured to capture a motion video for each sport using an image capture device in a corresponding area thereof, track and detect multiple athletes in the motion video based on a multi-target tracking detection model to determine multiple motion stages of each athlete and key state frame images of each motion stage, segment the agreed human body parts and rule marks in the key state frame images based on an image segmentation model, and generate a sports score of each athlete's violation status or valid score based on the segmentation result;

[0008] The performance storage module is used to store the student's sports performance.

[0009] In a preferred example, the multi-target tracking and detection model is a Yolox+Deepsort model, and the image segmentation model is a Unet model.

[0010] In a preferred example, the motion detection module is also used for the long jump event, using a multi-target tracking detection model to track and detect multiple athletes in the sports video to obtain a human body detection frame and a foot detection frame for each athlete, and based on the position and / or position change of the human body detection frame and the foot detection frame, determine multiple motion stages of each athlete and the key state frames of each motion stage.

[0011] In a preferred example, the motion detection module is also used to determine the take-off phase and take-off point state frame image, the flight phase and the highest point state frame image, the landing phase and the landing point state frame image of each athlete for the long jump event based on the position and / or position change of the human body detection frame and the foot detection frame, and use the image segmentation model to segment each state frame image to obtain the take-off point foot contour, landing point foot contour, highest point foot contour and safety line mark contour of each athlete. If the take-off point foot contour intersects with the safety line contour, it is determined to be an illegal line stepping. Otherwise, the long jump height and long jump distance are determined according to the take-off point foot contour, the highest point foot contour and the landing point foot contour to generate sports results of illegal line stepping or effective long jump distance and long jump distance.

[0012] In a preferred example, the movement stage also includes a retreat judgment stage; and the movement detection module is further used to determine that the current point is the landing point and enter the retreat judgment stage when the horizontal height of the human body detection frame in M ​​consecutive image frames in the landing stage does not change and is close to the horizontal height at the take-off stage for the long jump, and to determine that the current point is the landing point and enter the retreat judgment stage when the moving distance of the foot detection frame in γ consecutive image frames in the landing stage is less than a first threshold distance for the standing long jump. If the retreat distance of the human body detection frame or the foot detection frame in the retreat judgment stage reaches the second threshold distance or the outward expansion range of the foot detection frame toward the safety line reaches the threshold range, a retreat violation is determined and a movement score for a retreat violation is generated, where M≥2.

[0013] In a preferred example, the long jump event is a standing long jump or a running long jump, and the image acquisition device corresponding to the standing long jump event is arranged in front of the standing long jump area and the acquisition angle is opposite to the long jump direction, and the image acquisition device for the running long jump event is arranged on the side of the running long jump area and the acquisition angle is perpendicular to the long jump direction.

[0014] In a preferred embodiment, the motion detection module is further configured to determine a vertical distance between the highest point and a horizontal line based on the highest point foot contour and obtain a long jump height based on the vertical distance; and

[0015] Based on the take-off point foot contour, the landing point foot contour, and the safety line marker contour, the coordinates of the front point of the take-off point foot contour, the coordinates of the rear point of the landing point foot contour, and the safety line coordinates of the safety line marker contour in the image are obtained, and the long jump distance of the standing long jump or the running long jump is determined using the following formula based on the obtained coordinates:

[0016] D1=(X0+X1); D2=(Y0-Y1);

[0017] Y = H·tan(α+Δθ);

[0018]

[0019]

[0020] Among them, D1 is the long jump distance of the running long jump students, D2 is the long jump distance of the standing long jump students, X0 is the horizontal distance between the safety line and the image acquisition device, X1 is the horizontal distance between the last point of the foot contour at the landing point and the image acquisition device, Y0 is the vertical distance between the front point of the foot contour at the take-off point and the camera, Y1 is the vertical distance between the last point of the foot contour at the landing point and the image acquisition device, Y is the vertical distance between the target point and the image acquisition device, X is the horizontal distance between the target point and the image acquisition device, H is the height of the image acquisition device from the ground, α is the pitch angle of the image acquisition device, β is the horizontal field of view angle of the image acquisition device, θ is the vertical field of view angle, Δθ is the small step angle, and D min D is the vertical distance between the bottom edge of the image and the image acquisition device. max is the vertical distance between the top edge of the picture and the image acquisition device, x is the image horizontal coordinate of the target point, y is the image vertical coordinate of the target point, and h is the image height.

[0021] In a preferred example, the motion detection module is also used for long jump. When the moving speed of the human body detection frame in the take-off area reaches a threshold speed, the distance between the corresponding foot detection frame and the safety line is less than a third threshold distance, and the forward leaning angle of the human body in the human body detection frame is less than a threshold angle, it is determined that the current point is the take-off point and the take-off point status frame image is obtained, and the previously preset time period is the take-off stage. After that, the flight stage is entered, and the front and rear position changes of the human body detection frame in the flight stage are compared. The image frames with the highest corresponding height in the N human body detection frames are determined and obtained as the highest point status frame images and then the landing stage is entered. When the horizontal height of the human body detection frame in M ​​consecutive image frames in the landing stage does not change and is close to the horizontal height at the time of take-off, it is determined that the current point is the landing point and the landing point status frame image is obtained, and the landing stage ends.

[0022] In a preferred example, the motion detection module is further configured to determine, for standing long jump, that the current position is the take-off point when the distance between the foot detection frame and the safety line detection frame in the take-off area is less than a fourth threshold distance, obtain a take-off point state frame image, and enter a take-off phase; and when the corresponding human body detection frame intersects with the long jump area within a subsequent preset time period, enter a flight phase, compare the front-to-back position changes of the human body detection frames in the flight phase, determine and obtain the image frame with the highest corresponding height among N human body detection frames as the highest point state frame image, and then enter a landing phase; and when the moving distance of the foot detection frame in γ consecutive image frames in the landing phase is less than the first threshold distance, determine that the current position is the landing point, obtain a landing point state frame image, and end the landing phase, where γ ≥ 2;

[0023] The motion detection module is also used for standing long jump. If the movement speed of the human body detection frame exceeds a set threshold during the take-off phase and a preset time period before the take-off phase, a run-up violation is determined to generate a sports score of a run-up violation. Otherwise, whether there is a pad jump action is determined based on the position change of the foot detection frame during the take-off phase. When the up and down movement distance of the foot detection frame exceeds a fifth threshold distance, a pad jump violation is determined to generate a sports score of a pad jump violation. Once a violation is determined, the recognition of subsequent motion phases is stopped.

[0024] In a preferred embodiment, the system further comprises a big data center and a device management module;

[0025] The motion detection module obtains key state frame images of each motion stage of the sports event of the athlete to form an image frame sequence, and stores the image frame sequence in the performance storage module, and the big data center generates a motion trajectory line of the sports event of the athlete according to the image frame sequence stored in the performance storage module;

[0026] The equipment management module is respectively connected to the equipment set up on the playground to monitor the status information of each equipment. The equipment includes the image acquisition device, the sports operating console, the biometric recognition machine, the playground large screen, and the heart rate detection equipment. The heart rate detection equipment is assigned to students before they enter the playground to monitor the heart rate of the exercising students in real time and send the heart rate of the exercising students to the big data center. The big data center aligns the heart rate of the exercising students with the exercise trajectory of the exercising students in time.

[0027] The present application also discloses a motion detection method based on a campus intelligent playground system, wherein the system includes an image acquisition device provided at each sports area of ​​the playground; the method includes the following steps:

[0028] For each sport, an image acquisition device at a corresponding area is used to capture a sport video, and multiple athletes in the sport video are tracked and detected based on a multi-target tracking detection model to determine multiple sport stages of each athlete and key state frame images of each sport stage;

[0029] Segmenting the agreed human body parts and rule marks in the key state frame image based on an image segmentation model, and generating the violation status or valid sports score of each sports student according to the segmentation result;

[0030] The sports performance of the student is stored.

[0031] The embodiments of the present application include at least the following advantages and beneficial effects:

[0032] 1. Combining a multi-target tracking and detection model with an image segmentation model not only enables automatic detection and identification of athletic performance across different sports, but also allows for simultaneous tracking and detection of multiple athletes within the same sports area, yielding their performance data simultaneously. Furthermore, this unified automatic detection and identification approach allows for the acquisition of the same type of performance data across different sports, allowing for the integration of performance data across different sports. This integrated data allows for a timely and comprehensive analysis of each student's athletic performance.

[0033] 2. During the automatic detection and recognition process, multiple athletes in the sports video are tracked and detected based on a multi-target tracking detection model to determine multiple motion stages and key state frame images of each athlete in the corresponding motion stages, and the agreed human body parts and regular symbols in the key state frame images are segmented based on an image segmentation model. The sports scores of each athlete, including violations or valid scores, are generated based on the segmentation results. The advantages of this process include: first, the process can simultaneously track and detect multiple people in the same sports area and obtain the sports scores of multiple people at the same time, greatly improving the detection efficiency; especially for the long jump event, only one camera is needed to meet the automatic detection requirements, which greatly improves the detection efficiency of the long jump event while reducing equipment costs; second, the sports scores include not only valid scores but also violations, realizing the tracking and recording of violations, which provides certain guidance for subsequent teaching of students; third, the process involves few calculation parameters, simple calculation methods, and fast automatic detection and recognition speed.

[0034] 3. The aforementioned key state frame images can be stored to form an image frame sequence for each student. Based on the image frame sequence, the motion curve of each student's each movement can be directly obtained, and then, for example, a visual presentation can be performed.

[0035] The specification of this application records a large number of technical features, which are distributed in various technical solutions. If all possible combinations of technical features of this application (i.e., technical solutions) are to be listed, the specification will be too lengthy. In order to avoid this problem, the various technical features disclosed in the above-mentioned invention content of this application, the various technical features disclosed in the various embodiments and examples below, and the various technical features disclosed in the accompanying drawings can be freely combined with each other to form various new technical solutions (these technical solutions are all deemed to have been recorded in this specification), unless such a combination of technical features is technically infeasible. For example, in one example, feature A+B+C is disclosed, and in another example, feature A+B+D+E is disclosed. Features C and D are equivalent technical means that play the same role. Technically, only one of them can be used, and it is impossible to use them at the same time. Feature E can be technically combined with feature C. Then, the solution of A+B+C+D should not be considered as having been recorded because it is technically infeasible, while the solution of A+B+C+E should be considered as having been recorded. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a structural diagram of a campus smart playground system according to the first embodiment of the present application.

[0037] Figure 2 It is a flow chart of a motion detection method for a campus smart playground system according to the second embodiment of the present application.

[0038] Figure 3 This is a schematic diagram of the structure of a campus smart playground system according to an example of this application.

[0039] Figure 4 This is a layout diagram of an image acquisition device for a long jump event according to an example of the present application.

[0040] Figure 5 This is a flow chart of a motion detection method for a long jump event according to an example of the present application.

[0041] Figure 6 This is a flowchart of an automatic recognition process of sports performance based on an image segmentation model for a long jump event according to an example of the present application. DETAILED DESCRIPTION

[0042] In the following description, many technical details are provided to help readers better understand this application. However, those skilled in the art will understand that even without these technical details and various changes and modifications based on the following embodiments, the technical solutions claimed in this application can be implemented.

[0043] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0044] The first embodiment of the present application relates to a campus intelligent playground system, the structure of which is as follows: Figure 1 As shown, the campus intelligent playground system includes an image acquisition device, a motion detection module, and a performance storage module, which are installed in each sports area of ​​the playground. The motion detection module is used to capture a motion video of each sports event using the image acquisition device in the corresponding area. Using a multi-target tracking detection model, the module tracks and detects multiple athletes in the motion video to determine multiple motion phases and key state frame images of each motion phase. Using an image segmentation model, the module segments the designated human body parts and regular landmarks in the key state frame images. Based on the segmentation results, the module generates a sports score indicating each athlete's violation or valid score. The performance storage module stores the student's sports score.

[0045] The multi-target tracking detection model is pre-trained using an image data set marked with a predetermined detection frame, and the image segmentation model is pre-trained using an image data set marked with predetermined human body parts and regular identifiers. Optionally, the multi-target tracking detection model is a Yolox+Deepsort model, and the image segmentation model is a Unet model. Although only the combination of the Yolox+Deepsort model and the Unet model is listed, any combination of a multi-target tracking detection model and an image segmentation model that can realize the automatic recognition of sports performance in this application is within the scope of protection of this application.

[0046] In one embodiment, the motion detection module is also used for the long jump event. Based on the multi-target tracking detection model, the multiple athletes in the sports video are tracked and detected to obtain the human body detection frame and foot detection frame of each athlete. Based on the position and / or position change of the human body detection frame and the foot detection frame, the take-off phase and the take-off point state frame image, the flight phase and the highest point state frame image, the landing phase and the landing point state frame image of each athlete are determined. The image segmentation model is used to segment each state frame image to obtain the take-off point foot contour, the landing point foot contour, the highest point foot contour and the safety line mark contour of each athlete. If the take-off point foot contour intersects with the safety line contour, it is determined to be an illegal line stepping. Otherwise, the long jump height and long jump distance are determined according to the take-off point foot contour, the highest point foot contour and the landing point foot contour to generate sports results of illegal line stepping or effective long jump distance and long jump distance.

[0047] In an embodiment of the long jump event, the movement stage also includes a retreat judgment stage; and the movement detection module is further used to determine, for a running long jump, that the current landing point is the landing point and enter the retreat judgment stage when the horizontal height of the human body detection frame in M ​​consecutive image frames in the landing stage does not change and is close to the horizontal height at the take-off stage; and for a standing long jump, determine, for a standing long jump, that the current landing point is the landing point and enter the retreat judgment stage when the moving distance of the foot detection frame in γ consecutive image frames in the landing stage is less than a first threshold distance. If, in the retreat judgment stage, the retreat distance of the human body detection frame or the foot detection frame reaches a second threshold distance or the outward expansion range of the foot detection frame toward the safety line reaches a threshold range, a retreat violation is determined and a movement score for a retreat violation is generated, where M and γ ≥ 2.

[0048] Optionally, the long jump event can be a standing long jump or a running long jump, and the image acquisition device corresponding to the standing long jump event is set in front of the standing long jump area and the acquisition angle is opposite to the long jump direction, and the image acquisition device for the running long jump event is set to the side of the running long jump area and the acquisition angle is perpendicular to the long jump direction. In the scenario where the image acquisition device is arranged at a fixed angle and distance, the student's long jump height and long jump distance can be automatically calculated by the motion detection module through the following method: (1) determine the vertical distance from the highest point to the horizontal line based on the highest point foot contour and obtain the long jump height based on the vertical distance; (2) obtain the coordinates of the front point of the take-off point foot contour, the last point of the landing point foot contour and the safety line coordinates of the safety line marker contour in the image based on the take-off point foot contour, the landing point foot contour and the safety line marker contour, and determine the long jump distance of the standing long jump or the running long jump using the following formula based on the obtained coordinates:

[0049] D1=(X0+X1); D2=(Y0-Y1);

[0050] Y = H·tan(α+Δθ);

[0051]

[0052]

[0053] Among them, D1 is the long jump distance of the running long jump students, D2 is the long jump distance of the standing long jump students, X0 is the horizontal distance between the safety line and the image acquisition device, X1 is the horizontal distance between the last point of the foot contour at the landing point and the image acquisition device, Y0 is the vertical distance between the front point of the foot contour at the take-off point and the camera, Y1 is the vertical distance between the last point of the foot contour at the landing point and the image acquisition device, Y is the vertical distance between the target point and the image acquisition device, X is the horizontal distance between the target point and the image acquisition device, H is the height of the image acquisition device from the ground, α is the pitch angle of the image acquisition device, β is the horizontal field of view angle of the image acquisition device, θ is the vertical field of view angle, Δθ is the small step angle, and D min D is the vertical distance between the bottom edge of the image and the image acquisition device. max is the vertical distance from the top edge of the image to the image acquisition device, x is the image horizontal coordinate of the target point, y is the image vertical coordinate of the target point, and h is the image height. It should be noted that the target point corresponds to the point currently being calculated. For example, if the currently calculated point is the frontmost point of the foot, then that point is the target point; if the currently calculated point is the backmost point of the foot, then that point is the target point.

[0054] In an embodiment of the long jump project, the motion detection module can automatically detect its various motion stages, which are specifically described as follows: the motion detection module is optionally used for the long jump in a hurry. When the moving speed of the human body detection frame in the take-off area reaches a threshold speed, the distance between the corresponding foot detection frame and the safety line is less than a third threshold distance, and the forward leaning angle of the human body in the human body detection frame is less than a threshold angle, it is determined that the current point is the take-off point and the take-off point status frame image is obtained, and the previously preset time period is the take-off stage. After that, the flight stage is entered, and the front and rear position changes of the human body detection frame in the flight stage are compared. The image frame with the highest corresponding height in the N human body detection frames is determined and obtained as the highest point status frame image and then the landing stage is entered. When the horizontal height of the human body detection frame in M ​​consecutive image frames in the landing stage does not change and is close to the horizontal height at the take-off, it is determined that the current point is the landing point and the landing point status frame image is obtained, and the landing stage ends. The motion detection module is optionally used for standing long jump. When the distance between the foot detection frame and the safety line detection frame in the take-off area is less than a fourth threshold distance, the current point is determined to be the take-off point, and the take-off point status frame image is obtained, and the take-off phase is entered. When the corresponding human body detection frame intersects with the long jump area within a preset time period thereafter, the flight phase is entered, and the front and rear position changes of the human body detection frames in the flight phase are compared. The image frames with the highest corresponding heights in the N human body detection frames are determined and obtained as the highest point status frame images, and then the landing phase is entered. When the moving distance of the foot detection frame in γ consecutive image frames in the landing phase is less than the first threshold distance, the current point is determined to be the landing point, and the landing point status frame image is obtained, and the landing phase ends. The motion detection module is optionally also used for standing long jump. If the movement speed of the human body detection frame in the take-off stage and the previous preset time period exceeds a set threshold, the run-up violation is determined to generate a sports score of the run-up violation. Otherwise, it is determined whether there is a pad jump action based on the position change of the foot detection frame in the take-off stage. When the up and down movement distance of the foot detection frame exceeds the fifth threshold distance, the pad jump violation is determined to generate a sports score of the pad jump violation. Once a violation is determined, the identification of the subsequent motion stage is stopped, and N, M, γ≥2.

[0055] The motion detection module of the present embodiment can also determine the multiple motion phases of each sports student and the key state frame of each motion phase based on the position and / or position change of the detection frame of the different body parts of different sports events. In addition, this motion detection module can also determine the multiple motion phases of each sports student and the key state frame of each motion phase according to the demand agreement of violation judgment and effective result for different sports events, and can automatically detect the various motion phases of each sports event and automatically obtain the key state frame image of each motion phase. For example, for the high jump event, the take-off phase, the rising phase, the descending phase and the landing phase can be agreed upon. For another example, for long-distance running events (such as 800 meters, 1000 meters, etc.), the take-off phase, the running phase and the sprint phase can be agreed upon. Etc.

[0056] For example, in the embodiment of the pull-up project, the motion detection module is used to first use the YOLOX target detection model to detect the ascending and descending stages of the athlete and their key state frame images, and then use the image segmentation model to segment the human body parts, feet, hands, head and horizontal bar of the key state frame images to obtain the human body postures of each stage, and judge the completion status of the pull-ups of the athlete and count them according to the positional relationship between the human body postures and the horizontal bar in each stage.

[0057] For another example, in an embodiment of a long-distance running project, first, a number of image acquisition devices are deployed around the playground track (including a number of devices at the starting line, the finish line, and between the starting line and the finish line, and the devices at the starting line and the finish line are set just above the line and at the end of the line and the acquisition angle is toward the line), and the motion detection module is used to use the image acquisition device to capture motion video, and based on the multi-target tracking detection model, track and detect multiple athletes in the motion video to determine the starting phase, running phase, and sprint phase of each athlete, as well as the starting point state frame image, the finish line state frame image, and the running state frame image of each lap between the starting line and the finish line (for example, a circular running The image segmentation model is used to segment the feet and the starting line in the starting point state frame image, the human body and the finish line in the sprint state frame image, and the face in each lap running state frame image. The number of running laps is calculated according to the number of times the face is captured by the same camera between the starting line and the finish line. Whether each athlete has violated the rules and the violation situation is generated according to the positional relationship between the feet and the starting line in the starting phase. The starting time is determined according to the change in the positional relationship between the feet and the starting line in the starting phase. The time to cross the finish line is determined according to the change in the positional relationship between the human body and the finish line in the sprint phase. The running time is calculated according to the starting time and the time to cross the finish line.

[0058] Optionally, the campus intelligent playground system also includes a big data center and an equipment management module. The motion detection module obtains the key state frame image of each motion stage of the sports project of the sports student or a preset number of frame images before and after the key frame image to form an image frame sequence, and stores the image frame sequence in the performance storage module. The big data center generates the motion trajectory line of the sports project of the sports student based on the image frame sequence stored in the performance storage module. The equipment management module is respectively connected to the equipment set on the playground for monitoring the status information of each equipment. The equipment includes the image acquisition device, the sports console, the biometric recognition machine, the playground screen, and the heart rate detection equipment. The heart rate detection equipment is assigned to students before entering the playground to monitor the heart rate of the sports student in real time, and the heart rate of the sports student is sent to the big data center. The big data center aligns the heart rate of the sports student with the motion trajectory line of the sports student in time.

[0059] It should be noted that the various thresholds involved in this application can be set based on experience or obtained based on analysis of historical data.

[0060] The second embodiment of the present application relates to a motion detection method based on a campus intelligent playground system, which includes an image acquisition device set at each sports area of ​​the playground; the process of the motion detection method is as follows: Figure 2 As shown, the method includes the following steps:

[0061] In step 201, for each sports event, a sports video is captured using an image acquisition device at its corresponding area, and multiple sports students in the sports video are tracked and detected based on a multi-target tracking detection model to determine multiple sports stages of each sports student and key state frame images of each sports stage.

[0062] Then, enter step 202, segment the agreed human body parts and rule signs in the key state frame image determined in step 201 based on the image segmentation model, and generate the violation situation or valid sports score of each sports student according to the segmentation result.

[0063] Then, the process proceeds to step 203 to store the violation or valid sports performance obtained in step 202.

[0064] The multi-target tracking detection model in step 201 is pre-trained using an image dataset annotated with agreed detection frames, and the image segmentation model in step 202 is pre-trained using an image dataset annotated with agreed human body parts and regular identifiers. Optionally, the multi-target tracking detection model is a Yolox+Deepsort model, and the image segmentation model is a Unet model. Although only a combination of the Yolox+Deepsort model and the Unet model is listed, any combination of a multi-target tracking detection model and an image segmentation model that can implement the automatic recognition of athletic performance in this application is within the scope of protection of this application.

[0065] In one embodiment, step 201 and step 202 further include the following steps: for the long jump event, multiple athletes in the sports video are tracked and detected based on a multi-target tracking detection model to obtain a human body detection frame and a foot detection frame of each athlete; based on the position and / or position change of the human body detection frame and the foot detection frame, the take-off phase and the take-off point state frame image, the flight phase and the highest point state frame image, the landing phase and the landing point state frame image of each athlete are determined; and each state frame image is segmented using an image segmentation model to obtain the take-off point foot contour, the landing point foot contour, the highest point foot contour and the safety line mark contour of each athlete; if the take-off point foot contour intersects with the safety line contour, it is determined that the line is violated; otherwise, the long jump height and long jump distance are determined according to the take-off point foot contour, the highest point foot contour and the landing point foot contour to generate sports results of illegal line walking or effective long jump distance and long jump distance.

[0066] In an embodiment of the long jump event, the movement stage also includes a retreat judgment stage; and step 202 further includes the following steps: for the running long jump, when the horizontal height of the human body detection frame in M ​​consecutive image frames in the landing stage does not change and is close to the horizontal height at the take-off, it is determined that the current point is the landing point and the retreat judgment stage is entered; and for the standing long jump, when the moving distance of the foot detection frame in γ consecutive image frames in the landing stage is less than the first threshold distance, it is determined that the current point is the landing point and the retreat judgment stage is entered; if the retreat distance of the human body detection frame or the foot detection frame in the retreat judgment stage reaches the second threshold distance or the outward expansion range of the foot detection frame toward the safety line reaches the threshold range, it is determined that a retreat violation has occurred and a movement score for a retreat violation is generated, where M and γ are ≥ 2.

[0067] Optionally, the long jump event can be a standing long jump or a running long jump, and the image acquisition device corresponding to the standing long jump event is set in front of the standing long jump area and the acquisition angle is opposite to the long jump direction, and the image acquisition device for the running long jump event is set to the side of the running long jump area and the acquisition angle is perpendicular to the long jump direction. In the scenario where the image acquisition device is arranged at a fixed angle and distance, the student's long jump height and long jump distance can be automatically calculated by the motion detection module through the following method: (1) determine the vertical distance from the highest point to the horizontal line based on the highest point foot contour and obtain the long jump height based on the vertical distance; (2) obtain the coordinates of the front point of the take-off point foot contour, the last point of the landing point foot contour and the safety line coordinates of the safety line marker contour in the image based on the take-off point foot contour, the landing point foot contour and the safety line marker contour, and determine the long jump distance of the standing long jump or the running long jump using the following formula based on the obtained coordinates:

[0068] D1=(X0+X1); D2=(Y0-Y1);

[0069] Y = H·tan(α+Δθ);

[0070]

[0071]

[0072] Among them, D1 is the long jump distance of the running long jump students, D2 is the long jump distance of the standing long jump students, X0 is the horizontal distance between the safety line and the image acquisition device, X1 is the horizontal distance between the last point of the foot contour at the landing point and the image acquisition device, Y0 is the vertical distance between the front point of the foot contour at the take-off point and the camera, Y1 is the vertical distance between the last point of the foot contour at the landing point and the image acquisition device, Y is the vertical distance between the target point and the image acquisition device, X is the horizontal distance between the target point and the image acquisition device, H is the height of the image acquisition device from the ground, α is the pitch angle of the image acquisition device, β is the horizontal field of view angle of the image acquisition device, θ is the vertical field of view angle, Δθ is the small step angle, and D min D is the vertical distance between the bottom edge of the image and the image acquisition device. max is the vertical distance from the top edge of the image to the image acquisition device, x is the image horizontal coordinate of the target point, y is the image vertical coordinate of the target point, and h is the image height. It should be noted that the target point corresponds to the point currently being calculated. For example, if the currently calculated point is the frontmost point of the foot, then that point is the target point; if the currently calculated point is the backmost point of the foot, then that point is the target point.

[0073] In an embodiment of the long jump event, step 201 can be further implemented as follows: for a rush long jump, when the moving speed of the human body detection frame in the take-off area reaches a threshold speed, the distance between the corresponding foot detection frame and the safety line is less than a third threshold distance, and the forward leaning angle of the human body in the human body detection frame is less than a threshold angle, it is determined that the current point is the take-off point and the take-off point status frame image is obtained, and the previously preset time period is the take-off stage, and then the air stage is entered, and the front and rear position changes of the human body detection frame in the air stage are compared, and the image frames with the highest corresponding heights in the N human body detection frames are determined and obtained as the highest point status frame images, and then the landing stage is entered, and when the horizontal height of the human body detection frame in M ​​consecutive image frames in the landing stage does not change and is close to the horizontal height at the take-off point, it is determined that the current point is the landing point, and the landing point status frame image is obtained, and the landing stage ends. And step 201 can be further implemented as follows: for standing long jump, when the distance between the foot detection frame and the safety line detection frame in the take-off area is less than a fourth threshold distance, the current point is determined to be the take-off point, and the take-off point state frame image is obtained, and the take-off phase is entered. When the corresponding human body detection frame intersects with the long jump area within a preset time period, the flight phase is entered, and the front and rear position changes of the human body detection frame in the flight phase are compared. The image frame with the highest height corresponding to the N human body detection frames is determined and obtained as the highest point state frame image, and then the landing phase is entered. When the number of consecutive γ image frames in the landing phase is When the moving distance of the foot detection frame is less than the first threshold distance, it is determined that the current point is the landing point and the landing point status frame image is obtained, and the landing stage is ended; and if the moving speed of the human body detection frame in the take-off stage and the previous preset time period exceeds the set threshold, the run-up violation is determined to generate a sports score of the run-up violation, otherwise it is determined whether there is a pad jump action based on the position change of the foot detection frame in the take-off stage, and when the up and down moving distance of the foot detection frame exceeds the fifth threshold distance, the pad jump violation is determined to generate a sports score of the pad jump violation, wherein once a violation is determined, the recognition of the subsequent movement stage is stopped, N, M, γ≥2.

[0074] The motion detection method of the present embodiment can also determine the multiple motion stages of each sports student and the key state frame of each motion stage based on the position and / or position change of the detection frame of the different body parts agreed upon for different sports. In addition, the motion detection method can also agree on different multiple motion stages according to the demand of violation judgment and effective result for different sports, and can automatically detect the various motion stages of each sports and automatically obtain the key state frame image of each motion stage. For example, for the high jump project, the take-off stage, the rising stage, the descending stage and the landing stage can be agreed upon. For another example, for long-distance running projects (such as 800 meters, 1000 meters, etc.), the start stage, the running stage and the sprint stage can be agreed upon. Etc.

[0075] For example, in the embodiment of the pull-up project, step 201 and step 202 are further implemented as follows: first, based on the target detection model, the ascending and descending stages of the athlete, as well as the key state frame image of the highest point of the ascending stage and the key state frame image of the lowest point of the descending stage are detected, and then the human body parts, feet, hands, head and horizontal bar of the key state frame image are segmented using the image segmentation model to obtain the human body posture of each stage, and the completion status of the pull-up of the athlete is judged and counted according to the positional relationship between the human body posture and the horizontal bar in each stage.

[0076] For another example, in an embodiment of a long-distance running event, several image acquisition devices are first deployed around the playground track (including several devices at the starting line, the finish line, and between the starting line and the finish line. The devices at the starting line and the finish line are set directly above the line and at the end of the line, and the acquisition viewing angle is facing the direction of the line). And step 201 and step 202 are further implemented as follows: using an image acquisition device to capture a motion video, tracking and detecting multiple athletes in the motion video based on a multi-target tracking detection model to determine the starting phase, running phase and sprint phase of each athlete, as well as the starting point state frame image, the finishing line state frame image and the running state frame image of each lap between the starting line and the finish line (for example, a circular runway), and segmenting the feet and the starting line in the starting point state frame image, the human body part and the finish line in the finishing line state frame image, and the face of each lap of the running state frame image based on the image segmentation model, calculating the number of running laps according to the number of times the face is captured by the same camera between the starting line and the finish line, generating whether each athlete has violated the rules and the violation situation according to the positional relationship between the feet and the starting line in the starting phase, determining the starting time according to the change in the positional relationship between the feet and the starting line in the starting phase, determining the time to cross the finish line according to the change in the positional relationship between the human body part and the finish line in the sprint phase, and calculating the running time according to the starting time and the time to cross the finish line.

[0077] Optionally, the campus smart playground system also includes a big data center and a device management module. The motion detection method also includes the following steps A and B: Step A: Acquire key state frame images of each movement stage of the student's sport or a preset number of frame images before and after the key frame image to form an image frame sequence; Step B: Generate a motion trajectory line of the student's sport according to the image frame sequence.

[0078] Optionally, the motion detection method further includes the following steps: allocating heart rate detection equipment to students for them to wear before entering the playground to monitor the heart rate of the exercising students in real time; and aligning the heart rate of the exercising students with the motion trajectory line in time.

[0079] The first embodiment is a system embodiment corresponding to the present embodiment. The technical details in the first embodiment can be applied to the present embodiment, and the technical details in the present embodiment can also be applied to the first embodiment.

[0080] It should be noted that although the first and second embodiments only provide specific examples of three events: long jump, pull-ups and long-distance running, they cannot be used as a limitation on the scope of protection of this application. The motion detection module can be applied to any other sports that can be achieved based on the ideas of these two embodiments.

[0081] In order to better understand the technical solution of this application, the following is an example of a specific intelligent playground system and automatic detection of the long jump event. The details listed in this example are mainly for ease of understanding and are not intended to limit the scope of protection of this application. Figure 3 The example smart playground system shown. The system designs various sports intelligent detection modules according to the assessment standards of various sports, and especially designs a multi-person long jump detection module. This module realizes target tracking detection of multi-person long jump based on the multi-target tracking detection model, automatically identifies the various motion stages of the long jump target, and uses the Unet image segmentation model to obtain valid data on long jump distance and height, as well as invalid data for violations; secondly, based on the various sports data obtained, it generates student sports, examination and other performance reports. The system includes a motion detection module, a hardware equipment management module and a physical health performance management module. The motion detection module identifies and detects various sports of students and obtains motion data; the hardware equipment management module manages various hardware equipment in the playground; and the physical health performance management module manages students' various sports performance and physical fitness monitoring performance based on the motion data obtained by the motion detection module. The specific descriptions of each module are as follows:

[0082] 1. Arrange the image acquisition device: Figure 4 As shown, the image acquisition device corresponding to the standing long jump event is set in front of the standing long jump area and the acquisition angle is opposite to the long jump direction, and the image acquisition device for the emergency long jump event is set on the side of the emergency long jump area and the acquisition angle is perpendicular to the long jump direction.

[0083] Second, the Motion Detection Module: This module designs corresponding intelligent motion detection modules based on the physical education items and their assessment standards for the high school entrance exam. This module intelligently identifies and detects students' movements and generates student physical education assessment scores and campus movement trajectories based on the acquired motion data. For example, taking the long jump test as an example, an intelligent multi-person long jump detection method was designed based on the application requirements of large campuses.

[0084] Taking the long jump project as an example, the long jump detection module first uses the trained Yolox+Deepsort multi-target tracking model to detect the human body, feet and safety line in the video, and realizes the tracking of the human body and foot detection frames of the long jump object. The relative position relationship between the human body, foot detection frames and safety line detection frames is used to determine the stage of the long jump; secondly, the stored image frames of each stage of the long jump are used as the input of the Unet segmentation model to obtain the outline of the feet and safety line at each stage; finally, the long jump movement is analyzed and processed according to the detection and segmentation results to obtain the student's actual long jump distance and height. The specific implementation steps are as follows: Figure 5 shown.

[0085] 1. Model Training: Video streams of students performing long jump exercises from various schools were collected and captured as training data. The Yolox+Deepsort multi-object tracking and detection model's training data was annotated using the CVAT annotation tool, and the Unet segmentation model's training data was annotated using the LabelMe annotation tool. The prepared datasets were divided into an 80% training set and a 20% test set. The Yolox+Deepsort multi-object tracking and detection model (using the standard Yolox-Darknet53 algorithm) and the Unet segmentation model were trained separately.

[0086] 2. Based on the Yolox+Deepsort multi-target tracking and detection model, judge the various stages of the long jump process: ① First, set up camera 1 and camera 2 on the side and front of the long jump area. Students perform long jump in front of the safety line in different areas. Camera 1 and camera 2 respectively shoot the students' rapid long jump and standing long jump videos, and obtain the video stream data of the cameras through OpenCV. The actual long jump scene setting is shown in Figure 5 below. ② Each frame of the image is passed to the trained Yolox+Deepsort multi-target tracking and detection model in sequence to detect the position of the human body, feet and safety line in the video. The long jump area is processed using the safety line detection frame, and the rapid long jump and standing long jump and their take-off and long jump areas are divided. The human body movement trajectory in each area is tracked separately, and the long jump stage of each long jump object is judged based on the relative position relationship between the human body, foot detection frame and safety line detection frame. The specific judgment method is as follows:

[0087] 3. The judgment method for each stage of rapid long jump is as follows: a. Detect the moving speed of the human body detection frame in the take-off area of ​​rapid long jump. When the moving speed of the student's human body detection frame reaches the set threshold, the image frame is stored in a queue of corresponding length nα according to the detection frame ID (n seconds of data are stored according to actual conditions, and 1 second corresponds to α image frames); b. Determine whether the student has a tendency to long jump based on the distance between the foot detection frame and the safety line detection frame in the take-off area. When the distance between the student's foot detection frame and the safety line detection frame in the take-off area is less than the set threshold, the student is determined to be a pre-long jump target; c. Determine whether the pre-long jump target has taken off based on the forward inclination of the human body detection frame {the forward inclination of the human body in the take-off frame: the specific detection method can be the angle of the human body torso in the take-off frame or the angle of the line connecting the head and the feet, etc.}. When the forward inclination of the human body detection frame relative to the safety line detection frame reaches the set threshold, it is determined The pre-jump object takes off, and the current image frame is used as the take-off image frame and enters the air processing stage at the same time; d. Compare the front and back position changes of the long jump object's human body detection frame during the air process, select β (the default is 3) image frames with the highest corresponding height of the human body detection frame as the highest point image frame and enter the landing judgment stage; e. Determine whether the long jump object has landed based on the horizontal height change of the human body detection frame during the landing process. When the horizontal height of the human body detection frame in γ consecutive image frames does not change and is close to the horizontal height at the take-off, the current image frame is determined to be the landing moment image frame and enter the retreat judgment stage; f. Determine whether the long jump object retreats (i.e., moves toward the safety line) based on the position changes of the human body and foot detection frames in δ image frames after the landing moment. If the retreat distance of the human body or foot detection frame or the outward expansion range of the foot detection frame toward the safety line reaches the set threshold, it is determined that the long jump object has a retreat action.

[0088] 4. The judgment method for each stage of standing long jump is as follows: a. Detect the movement status of students in the standing long jump take-off area, exclude those who are not related to long jump, such as those sitting and resting, standing and talking, obtain the moving students in the standing long jump take-off area and store the image frames according to the detection frame ID; b. Determine whether the student has a tendency to long jump based on the distance between the foot detection frame and the safety line detection frame in the take-off area. When the distance between the student's foot detection frame and the safety line detection frame in the take-off area is less than the set threshold, the student is determined to be a pre-long jump target and enters the take-off judgment stage; c. Determine whether the pre-long jump target has taken off based on the degree of intersection between the human body detection frame and the long jump area. When the area of ​​the intersection area between the human body detection frame and the long jump area reaches the set threshold, the pre-long jump target is determined to have taken off and the current image frame is used as the take-off image frame to enter the run-up and pad jump detection stage at the same time; d. Determine whether the long jump target has a run-up based on the moving speed of the human body detection frame before take-off. When the moving speed of the long jump object's human body detection frame exceeds the set threshold, it is determined that there is a run-up behavior and the long jump stage judgment is stopped; if there is no run-up behavior, the long jump object is judged to have a pad jump action based on the position change of the foot detection frame in each image frame from pre-jump to take-off. When the up and down movement distance of the foot detection frame exceeds the set threshold, it is determined that the long jump object has a pad jump action and the long jump stage judgment is stopped. If there is no pad jump action, the long jump processing stage is entered; e. Compare the front and back position changes of the long jump object's human body detection frame during the pad jump process, select β (the default is 3) image frames with the highest corresponding height of the human body detection frame as the highest point image frame and enter the landing judgment stage; f. Determine whether the current long jump object has landed based on the position change of the foot detection frame during the landing process. When the moving distance of the foot detection frame in γ consecutive image frames is less than the set threshold, the current image frame is determined to be a landing image frame and enter the retreat judgment stage. g. Compare the position changes of the body and foot detection frames in the δ image frames after landing to determine whether the long jump subject has retreated. If the retreat distance of the body or foot detection frame, or the expansion range of the foot detection frame toward the safety line, reaches the set threshold, the long jump subject is determined to have retreated.

[0089] 5. Post-process and analyze the detection results based on the Unet image segmentation model

[0090] Based on the above detection and long jump stage judgment results, the identity of the long jump subject is confirmed and whether there are any illegal actions. The detection results of the long jump subjects without illegal actions are post-processed and analyzed, and the long jump distance and height are calculated to obtain the student's final long jump score. Figure 6The flowchart shown in the figure includes the following specific steps: ① Determine the identity of the long jump subject: Compare the long jump subject's facial image with the data in the campus face database to determine the long jump subject's identity. ② Illegal action judgment: Different criteria are used to judge illegal actions for different long jump types. For the long jump in a hurry, determine whether the long jump subject has taken a backward action. If so, directly output the invalid result and the reason for the invalidation. If not, perform post-processing and analysis of the test results. For the long jump in a standing position, determine whether the long jump subject has taken a run-up, a jump, or a backward action. If any of these illegal actions are present, directly output the invalid result and the corresponding reason for the invalidation. If not, perform post-processing and analysis of the test results. ③ Post-processing and analysis of detection results: The take-off image frame is fed into the trained Unet image segmentation model to obtain the outlines and coordinates of the foot and the safety line. If the foot in the take-off image frame intersects the safety line, the long jump subject is deemed to have stepped on the safety line, the long jump height and distance calculation is terminated, and the result and the reason for the invalidation are output. If there is no stepping on the safety line, the landing image frame and β stored image frames from the flight process are fed into the Unet segmentation model to obtain the foot outline at landing. The coordinates of the highest point of the body in the β flight image frames are compared to obtain the coordinates of the highest point. The student's actual long jump height and distance are calculated. In a scenario where cameras are arranged at a fixed angle and distance, the actual long jump height can be obtained by combining the vertical distance of the highest point of the long jump in the image with the horizontal line and an empirical ratio. The actual long jump distance is calculated based on the actual horizontal distance between the rearmost point of the foot and the camera at landing, the actual vertical distance between the frontmost point of the foot and the camera at take-off, and the actual vertical distance between the rearmost point of the foot and the camera at landing. The actual vertical distance between the image coordinate point and the camera is calculated according to the formulas described in the first and second embodiments above.

[0091] 6. Visualization of long jump results: Based on the above steps, the student's long jump trajectory is generated, including: ① Take-off line: parallel to the safety line and passing through the front point of the foot (the take-off line for long jump is the farthest end of the safety line); ② Long jump distance line: perpendicular to the take-off line and passing through the rearmost point of the foot; ③ Height line: perpendicular to the long jump distance line and passing through the highest point of the long jump; ④ Long jump trajectory line: an interpolated fitting curve passing through the take-off point, highest point, and landing point.

[0092] 3. Hardware equipment management module: This module includes sports consoles, network cameras, facial recognition machines, playground screens, heart rate monitoring equipment, intelligent recognition servers, etc., to enable managers to digitally manage the hardware equipment on the smart playground, including adding, deleting, timing power on and off, and remote management of equipment. When there is an abnormal connection between the equipment, it can promptly notify the management staff.

[0093] 4. Score storage module: This module includes regular sports scores, final special examination scores, physical fitness monitoring scores, health knowledge quiz scores, sports competition scores, etc., to manage and view the total physical health scores of the entire school, each grade, each class, and each student, as well as the specific scores of each individual item. Each score corresponds to specific sports information.

[0094] It should be noted that those skilled in the art should understand that the implementation functions of the modules shown in the above-mentioned embodiments of the campus intelligent playground system can be understood with reference to the relevant description of the motion detection method based on the campus intelligent playground system. The functions of the modules shown in the above-mentioned embodiments of the campus intelligent playground system can be implemented by programs (executable instructions) running on a processor, or by specific logic circuits. If the campus intelligent playground system described in the embodiment of the present application is implemented in the form of software function modules and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk. Thus, the embodiment of the present application is not limited to any specific combination of hardware and software.

[0095] It should be noted that in this patent application, relational terms such as first and second, etc., are used solely to distinguish one entity or operation from another, and do not necessarily require or imply any actual relationship or order between these entities or operations. Furthermore, the terms "comprise," "include," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. Without further limitation, an element specified by the phrase "comprising a" does not preclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element. In this patent application, reference to performing an action in accordance with an element means performing the action in accordance with at least that element, including two situations: performing the action in accordance with that element alone, and performing the action in accordance with that element and other elements. Expressions such as "plurality," "multiple times," and "many" include "two," "twice," "two kinds," and "more than two," "more than two times," and "more than two kinds."

[0096] All documents mentioned in this application are considered to be included in their entirety in the disclosure of this application so that they can be used as a basis for modification when necessary. In addition, it should be understood that the above description is only a preferred embodiment of this specification and is not intended to limit the scope of protection of this specification. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of this specification should be included in the scope of protection of one or more embodiments of this specification.

Claims

1. A campus intelligent playground system, characterized in that: include: An image acquisition device is installed at each sports area on the playground, wherein the image acquisition device for the emergency long jump event is located to the side of the emergency long jump area and the acquisition angle is perpendicular to the long jump direction; A motion detection module is configured to capture a motion video of a long jump event using an image capture device in a corresponding area thereof, track and detect multiple athletes in the motion video based on a multi-target tracking detection model, determine the current point to be the take-off point when the movement speed of a human body detection frame in the take-off area reaches a threshold speed, the distance between the corresponding foot detection frame and the safety line is less than a third threshold distance, and the forward lean angle of the human body in the human body detection frame is less than a threshold angle, and obtain a take-off point state frame image. The take-off stage is then entered, and the front-to-back position changes of the human body detection frames in the take-off stage are compared to determine and obtain image frames corresponding to the highest height in N human body detection frames as the highest point state frame images. The landing stage is then entered, and when the horizontal height of the human body detection frames in M ​​consecutive image frames in the landing stage remains unchanged and is close to the horizontal height at the time of take-off, determine the current point to be the landing point, obtain a landing point state frame image, and terminate the landing stage. The agreed human body parts and rule marks in each state frame image are segmented based on the image segmentation model, and a violation status or valid score of each athlete is generated based on the segmentation results. A performance storage module, used for storing the student's sports performance; Furthermore, the motion detection module determines the take-off phase and take-off point state frame image, the flight phase and highest point state frame image, the landing phase and landing point state frame image of each athlete based on the position and / or position change of the human body detection frame and the foot detection frame, and uses the image segmentation model to segment each state frame image to obtain the take-off point foot contour, the landing point foot contour, the highest point foot contour and the safety line marker contour of each athlete; if the take-off point foot contour intersects with the safety line marker contour, it is determined that the athlete has violated the line; otherwise, the long jump height and long jump distance are determined according to the take-off point foot contour, the highest point foot contour and the landing point foot contour, so as to generate the sports results of violating the line or valid long jump distance and long jump distance; The movement stage also includes a retreat judgment stage; and the movement detection module is further used to determine that the current landing point is the landing point and enter the retreat judgment stage when the horizontal height of the human body detection frame in M ​​consecutive image frames in the landing stage does not change and is close to the horizontal height at the take-off stage for the long jump, and to determine that the current landing point is the landing point and enter the retreat judgment stage when the moving distance of the foot detection frame in γ consecutive image frames in the landing stage is less than the first threshold distance for the standing long jump. If the retreat distance of the human body detection frame or the foot detection frame in the retreat judgment stage reaches the second threshold distance or the outward expansion range of the foot detection frame toward the safety line reaches the threshold range, a retreat violation is determined and a movement score for the retreat violation is generated, where M and γ are ≥ 2.

2. The campus intelligent playground system according to claim 1, characterized in that: The multi-target tracking detection model is a Yolox+Deepsort model, and the image segmentation model is a Unet model.

3. The campus intelligent playground system according to claim 1, characterized in that: The long jump event may also be a standing long jump, and the image acquisition device corresponding to the standing long jump event is arranged in front of the standing long jump area and the acquisition viewing angle is opposite to the long jump direction.

4. The campus intelligent playground system according to claim 1, characterized in that: The motion detection module is further configured to determine a vertical distance between the highest point and a horizontal line according to the highest point foot contour and obtain a long jump height according to the vertical distance; and Based on the take-off point foot contour, the landing point foot contour, and the safety line marker contour, the coordinates of the front point of the take-off point foot contour, the coordinates of the rear point of the landing point foot contour, and the safety line coordinates of the safety line marker contour in the image are obtained, and the long jump distance of the standing long jump or the running long jump is determined using the following formula based on the obtained coordinates: Among them, D1 is the long jump distance of the running long jump students, D2 is the long jump distance of the standing long jump students, X0 is the horizontal distance between the safety line and the image acquisition device, X1 is the horizontal distance between the last point of the foot contour at the landing point and the image acquisition device, Y0 is the vertical distance between the front point of the foot contour at the take-off point and the camera, Y1 is the vertical distance between the last point of the foot contour at the landing point and the image acquisition device, Y is the vertical distance between the target point and the image acquisition device, X is the horizontal distance between the target point and the image acquisition device, H is the height of the image acquisition device from the ground, α is the pitch angle of the image acquisition device, β is the horizontal field of view angle of the image acquisition device, θ is the vertical field of view angle, Δθ is the small step angle, and D min D is the vertical distance between the bottom edge of the image and the image acquisition device. max is the vertical distance between the top edge of the picture and the image acquisition device, x is the image horizontal coordinate of the target point, y is the image vertical coordinate of the target point, and h is the image height.

5. The campus intelligent playground system according to claim 1, characterized in that: The motion detection module is further configured to, for standing long jump, determine that the current position is the take-off point when the distance between the foot detection frame and the safety line detection frame in the take-off area is less than a fourth threshold distance, obtain a take-off point state frame image, and enter a take-off phase. Furthermore, when the corresponding human body detection frame intersects with the long jump area within a subsequent preset time period, enter a flight phase. Compare front-to-back position changes of the human body detection frames in the flight phase, determine and obtain an image frame corresponding to the highest height among N human body detection frames as a highest point state frame image, and then enter a landing phase. When the movement distance of the foot detection frame in γ consecutive image frames in the landing phase is less than the first threshold distance, determine that the current position is the landing point, obtain a landing point state frame image, and terminate the landing phase, where γ ≥ 2. The motion detection module is also used for standing long jump. If the movement speed of the human body detection frame exceeds a set threshold during the take-off phase and a preset time period before the take-off phase, a run-up violation is determined to generate a sports score of a run-up violation. Otherwise, whether there is a pad jump action is determined based on the position change of the foot detection frame during the take-off phase. When the up and down movement distance of the foot detection frame exceeds a fifth threshold distance, a pad jump violation is determined to generate a sports score of a pad jump violation. Once a violation is determined, the recognition of subsequent motion phases is stopped.

6. The campus intelligent playground system according to any one of claims 1 to 4, characterized in that: The system also includes a big data center and a device management module; The motion detection module obtains key state frame images of each motion stage of the sports event of the athlete to form an image frame sequence, and stores the image frame sequence in the performance storage module, and the big data center generates a motion trajectory line of the sports event of the athlete according to the image frame sequence stored in the performance storage module; The equipment management module is respectively connected to the equipment set up on the playground to monitor the status information of each equipment. The equipment includes the image acquisition device, the sports operating console, the biometric recognition machine, the playground large screen, and the heart rate detection equipment. The heart rate detection equipment is assigned to students before they enter the playground to monitor the heart rate of the exercising students in real time and send the heart rate of the exercising students to the big data center. The big data center aligns the heart rate of the exercising students with the exercise trajectory of the exercising students in time.

7. A motion detection method based on a campus intelligent playground system, characterized in that: The system includes an image acquisition device set at each sports event area on the playground, wherein the image acquisition device for the emergency long jump event is set on the side of the emergency long jump area and the acquisition angle is perpendicular to the long jump direction; The method comprises the following steps: For each sports event, a sports video is captured using an image capture device in a corresponding area thereof, and multiple sports students in the sports video are tracked and detected based on a multi-target tracking and detection model. When the movement speed of a human body detection frame in a take-off area reaches a threshold speed, the distance between a corresponding foot detection frame and a safety line is less than a third threshold distance, and the forward leaning angle of the human body in the human body detection frame is less than a threshold angle, it is determined that the current point is the take-off point and a take-off point state frame image is acquired. The previously preset time period is the take-off phase, and then the flight phase is entered. The front-to-back position changes of the human body detection frames in the flight phase are compared, and the image frames corresponding to the highest height in N human body detection frames are determined and acquired as the highest point state frame images. The landing phase is then entered. When the horizontal height of the human body detection frame in M ​​consecutive image frames in the landing phase does not change and is close to the horizontal height at the time of take-off, it is determined that the current point is the landing point and a landing point state frame image is acquired, and the landing phase ends. Segmenting the agreed human body parts and rule marks in each state frame image based on the image segmentation model, and generating the violation status or valid sports score of each sports student according to the segmentation result; storing the student's sports performance; Furthermore, the motion detection module determines the take-off phase and take-off point state frame image, the flight phase and highest point state frame image, the landing phase and landing point state frame image of each athlete based on the positions and / or position changes of the human body detection frame and the foot detection frame, and uses the image segmentation model to segment each state frame image to obtain the take-off point foot contour, the landing point foot contour, the highest point foot contour and the safety line marker contour of each athlete; if the take-off point foot contour intersects with the safety line marker contour, it is determined that the athlete has violated the line; otherwise, the long jump height and long jump distance are determined according to the take-off point foot contour, the highest point foot contour and the landing point foot contour, so as to generate the sports results of violating the line or valid long jump distance and long jump distance; The movement stage also includes a retreat judgment stage; and the movement detection module is further used to determine that the current point is the landing point and enter the retreat judgment stage when the horizontal height of the human body detection frame in M ​​consecutive image frames in the landing stage does not change and is close to the horizontal height at the take-off stage for the long jump, and to determine that the current point is the landing point and enter the retreat judgment stage when the moving distance of the foot detection frame in Y consecutive image frames in the landing stage is less than the first threshold distance for the standing long jump. If the retreat distance of the human body detection frame or the foot detection frame in the retreat judgment stage reaches the second threshold distance or the outward expansion range of the foot detection frame toward the safety line reaches the threshold range, a retreat violation is determined and a movement score for the retreat violation is generated, where M and Y ≥ 2.

Images