Tennis ball real-time landing detection method and device based on three-dimensional coordinates

Abstract

The invention discloses a tennis ball real-time landing detection method and device based on three-dimensional coordinates, and the method comprises the steps: collecting video frames in parallel through a plurality of cameras, and carrying out the time alignment to obtain a plurality of synchronous frames; recognizing two-dimensional pixel coordinates of the tennis ball in the multiple paths of synchronous frames, and constructing three-dimensional coordinates of the tennis ball in the space through a ray intersection method to obtain a continuous three-dimensional coordinate sequence; traversing the continuous three-dimensional coordinate sequence, and calculating the height value of each point; the height value of each point is compared with the height values of the front and back adjacent points, and whether the height of the middle point is the minimum height value or whether the trajectory has an inflection point is determined; if the height of the middle point is the height minimum value or the track has an inflection point, adding the middle point into a drop point candidate list; and selecting the candidate point with the lowest height value in the drop point candidate list as a drop point, and recording the coordinates of the drop point to obtain a drop point detection result. According to the application, the drop point of the tennis ball can be accurately identified by using the three-dimensional track, and misjudgment on the actual drop point is reduced.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of tennis landing detection, and in particular to a tennis real-time landing detection method and device based on three-dimensional coordinates. BACKGROUND

[0002] Tennis is a popular sport that attracts a large number of enthusiasts worldwide with its unique rules and high-intensity competition. In tennis matches, the decisions of referees are crucial, especially for determining whether the ball is in or out of the court. Traditional decision-making methods mainly rely on the eyesight and experience of referees, but manual observation often has certain subjective biases. For example, referees may make mistakes when judging the landing position due to different viewing angles or fatigue; at the same time, the penalty process usually takes 2-5 seconds, which may affect the smoothness of the game. More importantly, manual decisions may lead to inconsistent penalties due to differences in referees' understanding of the rules, especially in controversial decisions, which often need to be further confirmed through slow-motion playback, increasing the time cost and affecting the accuracy of the decision.

[0003] To ensure the fairness and accuracy of the game, the tennis community has introduced an automated decision-making system, which uses high-precision technology to track the trajectory of each ball in the game in real time, providing more accurate decision-making basis for referees, thereby improving the fairness and professionalism of the game. For example: the existing technology in "Research on Tennis Landing Positioning Based on Monocular Vision" proposes a tennis landing positioning system based on monocular vision, which can replace or supplement the high-priced professional landing tracking system that is difficult to deploy in low-cost environments.

[0004] Please refer to Figure 1 The core technology of the tennis landing positioning system based on monocular vision includes camera calibration, tracking landing, and landing positioning. In camera calibration, the corner points of the court lines are used as calibration feature points to solve the camera internal and external parameters and the lens distortion parameters, and the calibration feature points are optimized according to the re-projection error. In tracking landing, the background model and tracking algorithm are used to identify the foreground target, and the landing frame is determined according to the channel direction change of adjacent frames. In landing positioning, the lens distortion is removed from the landing in the picture, the projection model is applied, and the camera parameters are applied to map the landing coordinates to the three-dimensional world coordinate system.

[0005] This system monitors the tennis court through a monocular fixed camera. First, the image acquisition module continuously acquires video frames, and the image preprocessing module performs grayscale, denoising, and edge enhancement processing on the images. Then, the tennis detection module detects the position of moving objects (tennis balls) using frame difference method, and extracts the two-dimensional pixel coordinates of the ball in the image frame through the centroid positioning method. The trajectory extraction module linearly fits the pixel positions of the ball in consecutive frames to establish a two-dimensional trajectory.

[0006] The system judges whether the tennis ball lands or not and whether it is in the court by extracting the motion trend of the ball through the motion analysis module, judging the possible landing point frame according to the sharp decrease in speed or the extreme point in the y direction; then, the court line detection module detects the boundary line in the image based on Hough transform, and maps the landing point pixel coordinates to the actual court coordinate range; finally, the landing point judgment module compares the mapping result with the predefined court area, and outputs the penalty result.

[0007] The technology is mainly suitable for standard tennis court environment, and completes the identification and determination of the tennis landing point through monocular vision. It relies on stable image acquisition and image processing algorithm, and can realize a certain degree of automation in landing point detection. However, the technology is only based on two-dimensional coordinates in the landing point detection algorithm, the two-dimensional coordinate information is relatively single, which is easy to lead to misjudgment of the actual landing point. In addition, the two-dimensional coordinates are used for landing point detection without considering factors such as speed, acceleration and spatial boundary, when facing the two cases of netting and normal landing, since their speed and position change is relatively small, it is difficult to accurately distinguish them only by two-dimensional coordinates, and it is more likely to appear misjudgment and omission. SUMMARY

[0008] Therefore, the present application provides a tennis real-time landing detection method and device based on three-dimensional coordinates to solve the problem that the existing landing point detection algorithm is only based on two-dimensional coordinates, which is easy to misjudge the actual landing point.

[0009] In order to achieve the above purpose, the present application provides the following technical scheme:

[0010] In a first aspect, a tennis real-time landing detection method based on three-dimensional coordinates comprises:

[0011] Step 1: multiple cameras are used to collect video frames in parallel, and the multiple video frames are time-aligned to obtain multiple synchronous frames;

[0012] Step 2: a ball detection algorithm is used to identify the two-dimensional pixel coordinates of the tennis ball in the multiple synchronous frames, and a ray intersection method is used to construct the three-dimensional coordinates of the tennis ball in space to obtain a continuous three-dimensional coordinate sequence;

[0013] Step 3: the continuous three-dimensional coordinate sequence is traversed, and the height value of each point is calculated;

[0014] Step 4: the height value of each point is compared with the height values of the adjacent points before and after it, and it is determined whether the height of the intermediate point is the height minimum value or whether the trajectory has an inflection point;

[0015] Step 5: if the height of the intermediate point is the height minimum value or the trajectory has an inflection point, the intermediate point is added to the landing point candidate list;

[0016] Step 6: Select the candidate point with the lowest height value in the landing point candidate list as the landing point, and record the landing point coordinates to obtain the landing point detection result.

[0017] As a preferred, step 3 further comprises: performing second-order difference calculation on the continuous three-dimensional coordinate sequence, and calculating the acceleration maximum value, taking the acceleration maximum value point as auxiliary information of the landing point criterion.

[0018] As a preferred, it further comprises: measuring the spatial distance and time interval between adjacent two points, calculating the instantaneous speed of the tennis ball according to the spatial distance and time interval, and determining whether the landing point is the actual landing point according to the instantaneous speed.

[0019] As a preferred, the candidate points in the landing point candidate list are determined by calculating the average speed by the sliding window method and screening the speed mutation points according to the speed threshold.

[0020] As a preferred, it further comprises: using the spatial boundary conditions of the tennis court to spatially constrain the candidate points in the landing point candidate list.

[0021] As a preferred, for the candidate points in the landing point candidate list, the height minimum value, the acceleration maximum value, the instantaneous speed, the speed mutation and the spatial constraint are fused, and the confidence score is calculated, and the actual landing point is determined according to the confidence score.

[0022] As a preferred, when multiple landing points are detected in continuous multiple frames, the point with high consistency in spatial position, time and speed is selected as the final actual landing point.

[0023] The second aspect is a tennis real-time landing detection device based on three-dimensional coordinates, comprising:

[0024] A video frame acquisition module is configured to acquire video frames in parallel through multiple cameras and time-align the multiple video frames to obtain multiple synchronous frames.

[0025] A three-dimensional coordinate construction module is configured to identify two-dimensional pixel coordinates of the tennis ball in the multiple synchronous frames by using a ball detection algorithm, and construct three-dimensional coordinates of the tennis ball in space by using a ray intersection method to obtain a continuous three-dimensional coordinate sequence.

[0026] A height value calculation module is configured to traverse the continuous three-dimensional coordinate sequence and calculate the height value of each point.

[0027] A candidate point determination module is configured to compare the height value of each point with the height values of adjacent points before and after it, and determine whether the height of the intermediate point is a height minimum value or whether the trajectory has an inflection point; if the height of the intermediate point is a height minimum value or the trajectory has an inflection point, the intermediate point is added to the landing point candidate list.

[0028] The actual landing point determination module is configured to select the candidate point with the lowest height value in the landing point candidate list as the landing point, and record the landing point coordinates to obtain a landing point detection result.

[0029] In a third aspect, a computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the real-time tennis landing detection method based on three-dimensional coordinates when executing the computer program.

[0030] In a fourth aspect, a computer readable storage medium stores a computer program, and the computer program implements the steps of the real-time tennis landing detection method based on three-dimensional coordinates when executed by a processor.

[0031] Compared with the prior art, the present application has at least the following beneficial effects:

[0032] 1. The present application provides a real-time tennis landing detection method based on three-dimensional coordinates, which collects video frames in parallel through multiple cameras, and performs time alignment to obtain multiple synchronous frames; identifies the two-dimensional pixel coordinates of the tennis ball in the multiple synchronous frames, and constructs the three-dimensional coordinates of the tennis ball in space through the ray intersection method to obtain a continuous three-dimensional coordinate sequence; traverses the continuous three-dimensional coordinate sequence, calculates the height value of each point, determines the point with the minimum height value or judges whether the trajectory has an inflection point, and if the height of the intermediate point is the minimum height value or the trajectory has an inflection point, the intermediate point is added to the landing point candidate list; selects the candidate point with the lowest height value in the landing point candidate list as the landing point, and records the landing point coordinates to obtain a landing point detection result. The present application can accurately identify the landing point of the tennis ball based on the flight physical characteristics of the tennis ball using three-dimensional trajectory, and reduces the misjudgment of the actual landing point.

[0033] 2. For the candidate points in the landing point candidate list, the height minimum value, the acceleration maximum value, the instantaneous speed, the speed mutation and the space constraint are fused, and the confidence score is calculated, and the actual landing point is determined according to the confidence score. The present application fuses multiple dimensional features such as height, speed, acceleration and field space constraint, further judges the landing point through multiple frame data comparison and confidence score mechanism, and further reduces the misjudgment rate through multi-dimensional cross verification, significantly improves the accuracy and stability of the landing point detection, and greatly reduces the misjudgment and omission. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more intuitively illustrate the prior art and the present application, exemplary drawings are given below. It should be understood that the specific shapes, structures shown in the drawings should not be regarded as limiting conditions in the implementation of the present application; for example, based on the technical concepts disclosed in the present application and the exemplary drawings, those skilled in the art can easily make routine adjustments or further optimization to the increase / decrease / attribute division of certain units (components), specific shapes, positional relationships, connection modes, size ratio relationships, etc.

[0035] Figure 1 A single-vision-based tennis landing point positioning method flow chart provided by the prior art;

[0036] Figure 2 A three-dimensional coordinate-based tennis real-time landing detection method basic flow chart provided by the embodiment one of the present application;

[0037] Figure 3 A three-dimensional coordinate-based tennis real-time landing detection method judgment flow chart provided by the embodiment one of the present application. DETAILED DESCRIPTION

[0038] The present application will be further described in detail below with reference to the accompanying drawings.

[0039] In the description of the present application: unless otherwise specified, the meaning of "multiple" is two or more. The terms "first", "second", "third" and the like in the present application are intended to distinguish the objects referred to, and do not have a special meaning in the technical connotation aspect (for example, it should not be understood as an emphasis on importance or order, etc.). The expressions such as "include", "contain", "have" and the like also mean "not limited to" (certain units, components, materials, steps, etc.).

[0040] The terms such as "up", "down", "left", "right", "middle" and the like referred to in the present application are generally made for the purpose of indicating the relative position relationship for intuitive understanding with reference to the drawings, and are not an absolute limitation of the position relationship in the actual product.

[0041] Embodiment one

[0042] Please refer to Figure 2 and Figure 3 The embodiment provides a three-dimensional coordinate-based tennis real-time landing detection method, the core idea of which is based on the physical characteristics of the tennis in the flight process: after being hit, the height (y-axis, usually perpendicular to the direction of the field) of the tennis will first rise, start to descend after reaching the top, and finally rapidly decrease in height when contacting the ground, and the movement speed and direction of the ball will also change significantly after landing. Based on this physical law, the embodiment realizes accurate determination of the actual landing point by deeply analyzing the three-dimensional trajectory data of the tennis and comprehensively investigating various dynamic characteristics such as height change, speed change and acceleration change. Therefore, the method comprises:

[0043] S1: acquiring video frames in parallel through multiple cameras, and time-aligning the multiple video frames to obtain multiple synchronous frames;

[0044] Specifically, this embodiment employs multiple cameras to acquire video frames in parallel (each frame includes a precise timestamp), and uses a synchronization process to time-align the frames across all cameras, ensuring spatiotemporal consistency of the multi-camera data. The synchronized frame data is placed in a shared memory circular queue, and the GPU processing process reads the synchronized frames (frame data includes frame number, timestamp, and image data) from the queue. The shared memory structure includes write and read indexes, and a process lock is used to prevent contention for access by multiple processes.

[0045] This step of time-aligning multiple video frames specifically includes: the synchronization process monitors the shared memory of the two cameras in real time, compares the timestamps of each frame, and selects frames with timestamp intervals within a set allowable error range for time alignment. During the initial synchronization, strict time matching logic is used to ensure that the timestamps of the first paired frames from the two cameras are extremely close. The synchronized frame pairs are output to the processing queue in real time for use by subsequent 3D coordinate reconstruction algorithms. Compared with existing technologies, this embodiment, through an efficient shared memory mechanism and precise frame index and timestamp pairing, greatly improves the accuracy and real-time performance of multi-camera data synchronization, ensuring the efficiency and accuracy of subsequent 3D reconstruction and data processing.

[0046] S2: Use a ball detection algorithm to identify the two-dimensional pixel coordinates of the tennis ball in multi-channel synchronous frames, and construct the three-dimensional coordinates of the tennis ball in space through the ray intersection method to obtain a continuous three-dimensional coordinate sequence.

[0047] Specifically, constructing the 3D coordinates of a tennis ball in space using the ray intersection method includes: converting 2D pixel coordinates into rays in 3D space using pre-calibrated camera parameters; calculating the intersection points of each pair of rays from each camera; and selecting valid intersection points as 3D reconstruction points using a distance threshold (e.g., less than 50 mm), and recording the 3D coordinates of these points. If multiple candidate intersection points exist, the intersection point with the smallest ray distance is selected as the 3D reconstruction point for the current frame; if no valid intersection point exists, a placeholder (e.g., [-9999,-9999,-9999]) is used to indicate that there is no valid reconstruction data for the current frame. Finally, the valid 3D coordinates are output in real time to the subsequent processing module.

[0048] It should be noted that when calibrating the camera parameters (including intrinsic parameter matrix and distortion coefficients) for each camera in advance, precise camera calibration is performed by acquiring two-dimensional marker points on the site and known three-dimensional marker points.

[0049] S3: Traverse the continuous three-dimensional coordinate sequence and calculate the height value of each point;

[0050] Specifically, in this embodiment, the collected three-dimensional trajectory point sequence is traversed, and the height value of each point is calculated.

[0051] S4: comparing the height value of each point with the height values of the front and rear adjacent points, and determining whether the height of the middle point is a height minimum value or whether the trajectory has a turning point;

[0052] Specifically, the height of each point is compared with the heights of the front and rear points, and it is determined whether the point is a height minimum value or whether the trajectory has a turning point.

[0053] S5: if the height of the middle point is a height minimum value or the trajectory has a turning point, the middle point is added to the landing point candidate list;

[0054] Specifically, if it is found that the height value of a point is lower than a set threshold, and the height changes of the point and the front and rear points respectively show a downward and upward trend, i.e., the height from the previous frame to the frame is negative, and the height from the frame to the next frame is positive, it can be inferred that the point is the lowest point of the trajectory, and it is most likely the actual landing point of the ball. Therefore, the point can be added to the landing point candidate list.

[0055] S6: selecting the candidate point with the lowest height value in the landing point candidate list as the landing point, recording the landing point coordinates, and obtaining the landing point detection result.

[0056] The three-dimensional coordinate-based real-time tennis landing detection method provided in this embodiment can accurately identify the landing point of the tennis ball by using the three-dimensional trajectory, and reduces the misjudgment of the actual landing point.

[0057] In order to avoid occasional misjudgment, the three-dimensional coordinate-based real-time tennis landing detection method provided in this embodiment statistically analyzes the height changes of a plurality of consecutive frames, and only when the height suddenly rises after a continuous decrease and the lowest point height is lower than a threshold, the point is determined as the landing point.

[0058] The three-dimensional coordinate-based real-time tennis landing detection method provided in this embodiment also calculates the instantaneous speed of each point. The specific method is: measuring the spatial distance and time interval between adjacent two points, calculating the instantaneous speed of the tennis ball on the segment trajectory according to the spatial distance and time interval, and determining whether the landing point is the actual landing point according to the instantaneous speed. For example, if the speed before and after a certain moment suddenly changes obviously (for example: suddenly decreases from high speed to nearly zero), the point is most likely the actual landing point of the ball.

[0059] In order to improve the robustness of the determination, the three-dimensional coordinate-based real-time tennis landing detection method provided in this embodiment uses a sliding window method (such as the last 6 to 10 frames) to calculate the average speed, and combines a reasonable speed threshold (such as 10 to 220 km / h) to screen out the speed mutation point as the candidate landing point. That is, the candidate points in the landing point candidate list are determined by calculating the average speed by the sliding window method, and screening out the speed mutation point according to the speed threshold.

[0060] The tennis real-time landing detection method based on three-dimensional coordinates provided in this embodiment also analyzes the acceleration of the ball, and pays particular attention to the extreme value of the acceleration at the landing moment: through second-order difference calculation on the continuous three-dimensional coordinate sequence and calculation of the extreme value of the acceleration, the extreme value point of the acceleration is taken as auxiliary information of the landing point criterion, which can enhance the accuracy of the landing point determination.

[0061] The tennis real-time landing detection method based on three-dimensional coordinates provided in this embodiment also includes: using the spatial boundary conditions (such as the baseline, sideline and service area, etc.) of the tennis court to spatially constrain the candidate points in the landing point candidate list, that is, only the points located in the effective area of the court can be determined as the real landing point, and the points exceeding the boundary will be automatically excluded.

[0062] The tennis real-time landing detection method based on three-dimensional coordinates provided in this embodiment fuses the extreme value of the height, the extreme value of the acceleration, the instantaneous speed, the speed mutation and the spatial constraint for the candidate points in the landing point candidate list, and calculates the confidence score, and determines the actual landing point according to the confidence score. Specifically: for each candidate landing point, this embodiment will comprehensively consider its extreme value of the height, speed mutation, extreme value of the acceleration and spatial constraint, and calculate a confidence score in a multi-feature fusion manner (such as selecting the lowest point), and only the points with a confidence score exceeding a preset threshold will be finally determined as the actual landing point. If multiple candidate points simultaneously meet the conditions in the same period, the point with the highest confidence score will be selected as the final landing point.

[0063] The tennis real-time landing detection method based on three-dimensional coordinates provided in this embodiment selects the point with high consistency in spatial position, time and speed as the final actual landing point when multiple landing points are detected in continuous multiple frames.

[0064] The tennis real-time landing detection method based on three-dimensional coordinates provided in this embodiment automatically delays the output of the result for accidental misjudgment conditions, such as the ball being blocked or detection loss, and waits for the subsequent frame to supplement the data, so as to minimize the misjudgment and omission.

[0065] The tennis real-time landing detection method based on three-dimensional coordinates provided in this embodiment updates the current landing point state as soon as the actual landing point is detected, and records the spatial coordinates of the landing point, the detection frame number, the detection confidence, the ball speed and other information to the landing point history queue. Each landing point record contains the starting point (such as the service point or the last hitting point), the ending point (the current landing point), the detection time, the ball speed, the penalty state (such as In / Out / OffNet), the confidence and other detailed information.

[0066] The tennis real-time landing detection method based on three-dimensional coordinates provided in the embodiment is based on the flight physical characteristics of a tennis ball, fuses multi-dimensional characteristics of height, speed and acceleration and field space constraints, and further determines the landing point through multi-frame data comparison and confidence score mechanism. Compared with the traditional single feature detection, the multi-dimensional cross verification further reduces the misjudgment rate, significantly improves the accuracy and stability of the landing point detection, and greatly reduces the misjudgment and omission.

[0067] Embodiment Two

[0068] The tennis real-time landing detection device based on three-dimensional coordinates provided in the embodiment comprises:

[0069] The video frame acquisition module is configured to acquire video frames in parallel through multiple cameras and perform time alignment on the multiple video frames to obtain multiple synchronous frames.

[0070] The three-dimensional coordinate construction module is configured to identify two-dimensional pixel coordinates of the tennis ball in the multiple synchronous frames by using a ball detection algorithm, and construct three-dimensional coordinates of the tennis ball in space by using a ray intersection method to obtain a continuous three-dimensional coordinate sequence.

[0071] The height value calculation module is configured to traverse the continuous three-dimensional coordinate sequence and calculate a height value of each point.

[0072] The candidate point determination module is configured to compare the height value of each point with height values of adjacent points before and after the point, and determine whether the height of an intermediate point is a height minimum value or whether a trajectory has an inflection point. If the height of the intermediate point is a height minimum value or the trajectory has an inflection point, the intermediate point is added to a landing point candidate list.

[0073] The actual landing point determination module is configured to select a candidate point with the lowest height value in the landing point candidate list as a landing point, record a landing point coordinate, and obtain a landing point detection result.

[0074] The specific implementation content of each module in the tennis real-time landing detection device based on three-dimensional coordinates can be referred to the limitation of the tennis real-time landing detection method based on three-dimensional coordinates in the foregoing, and will not be described here.

[0075] Embodiment Three

[0076] The computer device provided in the embodiment comprises a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps of the tennis real-time landing detection method based on three-dimensional coordinates are implemented.

[0077] Embodiment Four

[0078] The embodiment provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize steps of a tennis real-time landing detection method based on three-dimensional coordinates.

[0079] The technical features of the above embodiments can be combined in any manner (as long as the combination of the technical features does not exist contradiction), in order to make the description simple, all possible combinations of the technical features in the above embodiments are not described; these embodiments which are not explicitly written should also be considered as the range of the present disclosure.

Claims

1. A method for real-time detection of a tennis ball landing based on three-dimensional coordinates, characterized in that, The method comprises the following steps: Step 1: acquiring video frames in parallel through multiple cameras, and time-aligning the multiple video frames to obtain multiple synchronous frames; Step 2: identifying the two-dimensional pixel coordinates of the tennis ball in the multiple synchronous frames by using a ball detection algorithm, and constructing the three-dimensional coordinates of the tennis ball in space by using a ray intersection method to obtain a continuous three-dimensional coordinate sequence; Step 3: traversing the continuous three-dimensional coordinate sequence and calculating the height value of each point; Step 4: comparing the height value of each point with the height values of the adjacent points before and after the point, and determining whether the height of the intermediate point is a height minimum value or whether the trajectory has a turning point; Step 5: if the height of the intermediate point is a height minimum value or the trajectory has a turning point, the intermediate point is added to a landing point candidate list; Step 6: selecting the candidate point with the lowest height value in the landing point candidate list as the landing point, and recording the landing point coordinates to obtain a landing point detection result.

2. The method of claim 1, wherein the method is based on three-dimensional coordinates. Step 3 further comprises: performing second-order difference calculation on the continuous three-dimensional coordinate sequence, and calculating an acceleration maximum value, and taking the acceleration maximum value point as auxiliary information for landing point determination.

3. The method of claim 2, wherein the three-dimensional coordinate-based real-time tennis ball landing detection method is characterized by, Further comprising: measuring the spatial distance and time interval between two adjacent points, calculating the instantaneous speed of the tennis ball according to the spatial distance and the time interval, and determining whether the landing point is an actual landing point according to the instantaneous speed.

4. The method of claim 3, wherein the method is based on three-dimensional coordinates. The candidate points in the landing point candidate list are determined by calculating the average speed by using a sliding window method and screening out the speed mutation points according to a speed threshold.

5. The method of claim 4, wherein the three-dimensional coordinate-based real-time tennis ball landing detection method is characterized by, Further comprising: spatially constraining the candidate points in the landing point candidate list by using the spatial boundary conditions of the tennis court.

6. The method of claim 5, wherein the three-dimensional coordinate-based real-time tennis ball landing detection method is characterized by, For the candidate points in the landing point candidate list, the height minimum value, the acceleration maximum value, the instantaneous speed, the speed mutation and the spatial constraint are fused, and a confidence score is calculated, and the actual landing point is determined according to the confidence score.

7. The method of claim 1, wherein the method is a three-dimensional coordinate-based real-time tennis ball landing detection method, characterized by, When multiple landing points are detected in continuous multiple frames, the point with high consistency in spatial position, time and speed is selected as the final actual landing point.

8. A real-time landing detection device for a tennis ball based on three-dimensional coordinates, characterized by, The method comprises the following steps: a video frame acquisition module for acquiring video frames in parallel through multiple cameras, and time-aligning the multiple video frames to obtain multiple synchronous frames; a three-dimensional coordinate construction module for identifying the two-dimensional pixel coordinates of the tennis ball in the multiple synchronous frames by using a ball detection algorithm, and constructing the three-dimensional coordinates of the tennis ball in space by using a ray intersection method to obtain a continuous three-dimensional coordinate sequence; a height value calculation module for traversing the continuous three-dimensional coordinate sequence and calculating the height value of each point; a candidate point determination module for comparing the height value of each point with the height values of the adjacent points before and after the point, and determining whether the height of the intermediate point is a height minimum value or whether the trajectory has a turning point; if the height of the intermediate point is a height minimum value or the trajectory has a turning point, the intermediate point is added to a landing point candidate list; an actual landing point determination module for selecting the candidate point with the lowest height value in the landing point candidate list as the landing point, and recording the landing point coordinates to obtain a landing point detection result. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to realize the steps of the method in any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the method in any one of claims 1 to 7.

Images