A method for measuring the run-up speed of long jump based on machine vision

Abstract

The invention discloses a method for measuring the run-up speed of a long jump based on machine vision, comprising the following steps: S1. determining the placement position of a camera and shooting parameters; S2. determining the position of a marker; S3. preprocessing of a video image; and S4. measuring the run-up speed. This method is easy to operate, low in cost, strong in scalability, and improves the robustness of speed measurement.

Description

Technical Field

[0001] The invention relates to the technical field of machine vision, in particular to a method for measuring the run-up speed of a long jump based on machine vision. Background Art

[0002] Long jump is one of the most common track and field sports. The technical points of long jump are generally divided into four stages: run-up, take-off, take-off and landing. Among them, the run-up speed is one of the main factors affecting the long jump results. Athletes can only get ideal long jump results if they obtain a reasonable run-up speed according to their own abilities. The run-up speed value is generally measured by professional equipment such as high-speed video analysis or infrared photoelectric timing gates. Due to the expensive, complex and inconvenient installation of the equipment, it is difficult to popularize it in the teaching scenarios of long jump courses in colleges and universities. Therefore, it is very necessary to study a low-cost run-up speed measurement solution. Summary of the invention

[0003] The purpose of the present invention is to provide a method for measuring the run-up speed of a long jump based on machine vision in view of the shortcomings of the prior art. The method is easy to operate, low in cost, strong in scalability, and improves the robustness of speed measurement.

[0004] The technical solution for achieving the purpose of the present invention is:

[0005] A method for measuring the run-up speed of a long jump based on machine vision comprises the following steps:

[0006] S1. Determine the camera placement and shooting parameters: Determine the camera placement and the camera shooting parameters during the long jump according to the characteristics of the long jump scene. The camera is placed at a vertical distance of D meters from the sideline of the run-up area and a horizontal vertical distance of d1+d2 meters from the take-off line. The camera is used to obtain the video sequence of the run-up process.

[0007] S2. Determine the position of the markers: Place markers on the sidelines of the run-up area and on both edges of the run-up area. 11 、p 12 、p 21 、p 22 、p 31 、p 32 、p 41 、p 42 , p 11 With p 12 The connection line is L1, p 21 With p 22 The connection line is L2, p 31 With p 32 The connection is L3, p 41 With p 42The connecting line is L4, L1, L2, L3, L4 are all perpendicular to the edge line of the run-up area, the distances between L1 and L2 and between L3 and L4 are both d1, and the perpendicular distances between L2 and L3 and the vertical line of the camera are both d2;

[0008] S3. Video image preprocessing: Place the camera and marker according to steps S1 and S2, record the running process video, extract the first frame image in the video sequence, and manually mark the marker in the image to obtain the coordinates of the marker in the image;

[0009] S4. Measure the running speed: bbox is the human body bounding box, p is the bottom midpoint of the bounding box, and the specific speed measurement process is as follows:

[0010] S4-1: Output the human bounding box bbox. Use the trained target detection algorithm YOLOv5 model to detect the human body in the long jump run-up video to obtain the athlete's position information in the video frame. In order to avoid interference from other people, use the Deepsort target tracking algorithm to predict the athlete's position information in the next frame. Combine the target detection algorithm and the Deepsort target tracking algorithm to achieve athlete tracking and positioning. Use the trained YOLOv5-Deepsort target tracking detection model to perform human body detection on the input video stream frame by frame. When the human body is detected and the position information of the human body bounding box is in the run-up area, the athlete's position information in the video frame is output in real time, that is, the YOLOv5-Deepsort target tracking detection algorithm detects and outputs the rectangular bounding box of the human body. The output bounding box information is (x min ,y min , x max ,y max ), the coordinates of the bottom center point p of the player's position bounding box are calculated as:

[0011]

[0012] S4-2: In line L1, let p 11 、p 12 The coordinates of the two points are (x 11 ,y 11 )、(x 12 ,y 12 ), find the point p through the coordinates of the two points 11 、p 12 The equation of the straight line is:

[0013] Ax+By+C=0 (2)

[0014] Where A = y 12 -y 11 , B = x 11 -x 12 , C = x12 *y 11 -x 11 *y 12 , similarly, find the point p 21 、p 22 The L2 straight line equation of

[0015] S4-3: Take the bottom center point of the bounding box p(x p ,y p ) as the detection point, and determine which side of the straight line point p is on according to formula (3);

[0016] D=A*x p +B*y p +c (3)

[0017] The deflection direction of the straight line L1 is defined by p 11 Point to p 12 When D<0, point p is on the right side of line L1; when D=0, point p is on line L1; when D>0, point p is on the left side of line L1; for line L1, when D≤0 is detected, it means that point p passes through line L1, and the current video stream frame number F is recorded. s Similarly, when passing through L2, record the current frame number F e ;

[0018] S4-4: Calculate the average speed of the athlete when passing through the straight lines L1 and L2:

[0019]

[0020] where d l1,2 is the actual distance between lines L1 and L2, FPS is the frame rate of the input video, is the actual time that point p passes through the two straight lines L1 and L2;

[0021] S4-5: Repeat steps S4-2 to S4-4 to obtain the frame numbers of the athlete passing through the straight lines L3 and L4, respectively. s ′,F e ′, average speed v L3,4 In order to verify and analyze the influence of the athlete's acceleration in a certain distance on the long jump performance, the athlete's acceleration in this area is calculated as:

[0022]

[0023] where S is the athlete's displacement, v L1,2 and v L3,4 is the final measured approach speed.

[0024] The camera shooting parameters in step S1 include the resolution, frame rate and storage format used when shooting the video.

[0025] The key points of this technical solution are:

[0026] (1) To avoid interference from other people during the run-up phase, the YOLOv5-DeepSORT target tracking algorithm is used to track the athletes in the run-up area in real time, and people outside the run-up area are not tracked. The second purpose of using the target tracking algorithm to track the run-up is to lay a foundation for the subsequent analysis of the run-up trajectory, which fully demonstrates the scalability of this speed measurement method.

[0027] (2) A virtual line running speed measurement model is constructed based on the bounding box output by the target tracking algorithm, and the average speed of the runner within a certain distance can be calculated.

[0028] When this method uses a virtual line speed measurement model for speed measurement, it tracks the athlete through the YOLOv5-DeepSORT target tracking and detection algorithm to avoid interference from other external personnel and improve the robustness of speed measurement. In addition, this method measures the run-up speed based on machine vision, which can reduce the cost of speed measurement while improving the scalability of speed measurement. For example, the YOLOv5-DeepSORT algorithm can perform motion analysis such as trajectory tracking during the run-up phase of the athlete. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flow chart of an embodiment;

[0030] Figure 2 A diagram showing the placement of cameras and markers in the embodiment;

[0031] Figure 3 This is a video image preprocessing result diagram in the embodiment;

[0032] Figure 4 It is a virtual linear velocity measurement model diagram in the embodiment;

[0033] Figure 5 The speed measurement algorithm in this embodiment calculates the athlete's running speed result diagram. DETAILED DESCRIPTION

[0034] The content of the present invention is further described below in conjunction with the drawings and embodiments, but the present invention is not limited thereto.

[0035] Example:

[0036] Reference Figure 1 A method for measuring the run-up speed of a long jump based on machine vision comprises the following steps:

[0037] S1. Determine the camera placement and shooting parameters: Determine the camera placement and the camera shooting parameters during the long jump according to the characteristics of the long jump scene. The camera is placed at a vertical distance of D meters from the edge of the run-up area and a horizontal vertical distance of d1+d2 meters from the take-off line. The camera is used to obtain the video sequence of the run-up process. The camera shooting parameters include the resolution, frame rate, and storage format used when shooting the video;

[0038] S2. Determine the location of the marker: Figure 2 As shown, place markers on the sideline of the run-up area and place markers on both edges of the run-up area. 11 、p 12 、p 21 、p 22 、p 31 、p 32 、p 41 、p 42 , p 11 With p 12 The connection line is L1, p 21 With p 22 The connection line is L2, p 31 With p 32 The connection is L3, p 41 With p 42 The connecting line is L4, L1, L2, L3, L4 are all perpendicular to the edge line of the run-up area, the distances between L1 and L2 and between L3 and L4 are both d1, and the perpendicular distances between L2 and L3 and the vertical line of the camera are both d2;

[0039] S3. Video image preprocessing: Place the camera and markers according to steps S1 and S2, record the running process video, extract the first frame image in the video sequence, and manually mark the markers in the image, such as Figure 3 As shown, the coordinates of the marker in the image are obtained;

[0040] S4.Measure the running speed: Figure 4 As shown, bbox is the human body bounding box, p is the bottom midpoint of the bounding box, and the specific speed measurement process is as follows:

[0041] S4-1: Output the human bounding box bbox. Use the trained target detection algorithm YOLOv5 model to detect the human body in the long jump run-up video to obtain the athlete's position information in the video frame. In order to avoid interference from other people, use the Deepsort target tracking algorithm to predict the athlete's position information in the next frame. Combine the target detection algorithm and the Deepsort target tracking algorithm to achieve athlete tracking and positioning. Use the trained YOLOv5-Deepsort target tracking detection model to perform human body detection on the input video stream frame by frame. When the human body is detected and the position information of the human body bounding box is in the run-up area, the athlete's position information in the video frame is output in real time, that is, the YOLOv5-Deepsort target tracking detection algorithm detects and outputs the rectangular bounding box of the human body. The output bounding box information is (x min ,y min , x max ,y max ), the coordinates of the bottom center point p of the player's position bounding box are calculated as:

[0042]

[0043] S4-2: In line L1, let p 11 、p 12 The coordinates of the two points are (x 11 ,y 11 )、(x 12 ,y 12 ), find the point p from the coordinates of the two points 11 、p 12 The equation of the straight line is:

[0044] Ax+By+C=0 (2)

[0045] Where A = y 12 -y 11 , B = x 11 -x 12 , C = x 12 *y 11 -x 11 *y 12 , similarly, find the point p 21 、p 22 The L2 straight line equation of

[0046] S4-3: Take the bottom center point of the bounding box p(x p ,y p ) as the detection point, and determine which side of the straight line point p is on according to formula (3);

[0047] D=A*x p +B*y p +C (3)

[0048] The deflection direction of the straight line L1 is defined by p 11 Point to p 12 When D<0, point p is on the right side of line L1; when D=0, point p is on line L1; when D>0, point p is on the left side of line L1; for line L1, when D≤0 is detected, it means that point p passes through line L1, and the current video stream frame number F is recorded. s Similarly, when passing through L2, record the current frame number F e ;

[0049] S4-4: Calculate the average speed of the athlete when passing through the straight lines L1 and L2:

[0050]

[0051] where d l1,2 is the actual distance between lines L1 and L2, FPS is the frame rate of the input video, is the actual time that point p passes through the two straight lines L1 and L2;

[0052] S4-5: Repeat steps S4-2 to S4-4 to obtain the frame numbers of the athlete passing through the straight lines L3 and L4, respectively. s ′,F e ′, average speed v L3,4 In order to verify and analyze the influence of the athlete's acceleration in a certain distance on the long jump performance, the athlete's acceleration in this area is calculated as:

[0053]

[0054] Where S is the displacement of the athlete, such as Figure 2 As shown, where v L1,2 and v L3,4 is the final measured approach speed.

[0055] Simulation experiment: Step 4 can be used to obtain the average speed of the athlete during the long jump. The average speed of the athlete passing through the virtual lines L1 and L2 is v1, and the speed passing through the virtual lines L3 and L4 is v2. The values ​​of v1 and v2 are calculated using the virtual line speed measurement algorithm, as shown in Figure 5 shown.

Claims

1. A method for measuring the run-up speed of a long jump based on machine vision, characterized in that: The steps include: S1. Determine camera placement and shooting parameters: Determine the camera placement and shooting parameters for the long jump based on the characteristics of the long jump scenario. The camera is placed at a vertical distance of D meters from the edge of the run-up area and a horizontal vertical distance of d1 + d2 meters from the take-off line. The camera is used to capture the video sequence of the run-up process. S2. Determine the position of the markers: Place markers on the edge of the run-up area and place markers on both edges of the run-up area. 11 、p 12 、p 21 、p 22 、p 31 、p 32 、p 41 、p 42 , p 11 With p 12 The connection line is L1, p 21 With p 22 The connection line is L2, p 31 With p 32 The connection is L3, p 41 With p 42 The connecting line is L4, L1, L2, L3, and L4 are all perpendicular to the edge of the run-up area, the distances between L1 and L2 and between L3 and L4 are both d1, and the perpendicular distances between L2 and L3 and the camera mid-perpendicular line are both d2; S3. Video image preprocessing: Place the camera and markers according to steps S1 and S2, record the run-up process, extract the first frame of the video sequence, and manually mark the markers in the image to obtain the coordinates of the markers in the image; S4. Measure the running speed: bbox is the bounding box of the human body, and p is the bottom midpoint of the bounding box. The specific speed measurement process is as follows: S4-1: Output the human bounding box bbox. Use the trained target detection algorithm YOLOv5 model to detect the human body in the long jump run-up video and obtain the athlete's position information in the video frame. In order to avoid interference from other people, use the Deepsort target tracking algorithm to predict the athlete's position information in the next frame. Combine the target detection algorithm and the Deepsort target tracking algorithm to achieve athlete tracking and positioning. Use the trained YOLOv5-Deepsort target tracking detection model to perform human body detection on the input video stream frame by frame. When the human body is detected and the position information of the human body bounding box is in the run-up area, the athlete's position information in the video frame is output in real time. That is, the YOLOv5-Deepsort target tracking detection algorithm detects and outputs the rectangular bounding box of the human body. The output bounding box information is (x min ,y min , x max ,y max ), the coordinates of the bottom center point p of the player's position bounding box are calculated as: S4-2: In line L1, let p 11 、p 12 The coordinates of the two points are (x 11 ,y 11 )、(x 12 ,y 12 ), find the point p from the coordinates of the two points 11 、p 12 The equation of the line is: Ax+By+C=0 (2) Where A = y 12 -y 11 , B=x 11 -x 12 , C=x 12 *y 11 -x 11 *y 12 , similarly find the point p 21 、p 22 The L2 straight line equation; S4-3: Take the bottom center point of the bounding box p(x p ,y p ) as the detection point, and determine which side of the line point p is on according to formula (3); D=A*x p +B*y p +C (3) The deflection direction of the straight line L1 is defined by p 11 Point to p 12 When D < 0, point p is on the right side of line L1; when D = 0, point p is on line L1; when D > 0, point p is on the left side of line L1; for line L1, when D ≤ 0 is detected, it means that point p passes through line L1, and the current video stream frame number F is recorded. s Similarly, when passing through L2, record the current frame number F e ; S4-4: Calculate the average speed of the athlete when passing through straight lines L1 and L2: where d l1,2 is the actual distance between the lines L1 and L2, and FPS is the frame rate of the input video. is the actual time that point p passes through the two straight lines L1 and L2; S4-5: Repeat steps S4-2 to S4-4 to obtain the number of frames of the athlete passing through the straight lines L3 and L4, respectively. s , F′ e , average speed v L3,4 In order to verify and analyze the influence of the athlete's acceleration in a certain distance and the long jump performance, the athlete's acceleration in this area is calculated as: where S is the displacement of the athlete, v L1,2 and v L3,4 is the final measured approach speed.

2. The method for measuring the run-up speed of a long jump based on machine vision according to claim 1, wherein: The camera shooting parameters in step S1 include the resolution, frame rate and storage format used when shooting the video.

Images