Track race evaluation method based on machine vision
By collecting multi-frame images to recognize keyframes, combining position relationship and time, the problem of not being able to track the entire process in track racing results evaluation is solved, and the full-process state evaluation and autonomous training of the movement process are realized.
Patent Information
- Application Number
- CN202311852812.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-29
- Publication Date
- 2025-07-01
AI Technical Summary
The existing technology cannot track the full-process sports process in track and track performance evaluation, especially for middle- and long-distance running trainers, who can only obtain final results and cannot meet the full-process status assessment needs of individuals.
By collecting multi-frame images, identifying keyframes and analyzing the positional relationship between the target individual and the identifier, combining motion distance and time, the evaluation of the motion process is achieved.
It realizes the full-process state evaluation of the exercise process, supports independent training, can correct the jump results, and provides segmented sports data reference.
Abstract
Description
Technical Field
[0001] The present invention relates to a track and field evaluation method based on machine vision. Background Art
[0002] For example, the performance evaluation of track and field events such as 50 meters and 100 meters is becoming increasingly intelligent. The test method based on machine vision is also one of them. It has many advantages. It does not require wearing equipment for assistance, nor does it need to bury sensors underground. The layout cost is relatively low. However, in the prior art, the performance evaluation is mainly based on the following method: obtaining the starting time of the target individual on the starting side, obtaining the arrival time of the target individual on the ending side, and then calculating the performance of the moving individual based on the moving distance. However, for middle and long-distance runners, especially trainers, only obtaining the final result and not being able to track the entire state during the movement process obviously cannot fully meet the needs of all moving individuals and urgently needs to be improved. Summary of the Invention
[0003] To solve the technical problems in the background art, a track and field evaluation method based on machine vision of the present invention includes: collecting multiple frames of first images, identifying a first key frame from the multiple frames of first images, and there is a preset positional relationship between the target individual and the first identifier in the first key frame; collecting multiple frames of second images, identifying a second key frame from the multiple frames of second images, and there is a preset positional relationship between the target individual and the second identifier in the second key frame; evaluating the performance of the target individual during the movement process between the first identifier and the second identifier according to the moving distance of the target individual analyzed according to a preset rule and the acquisition times of the first key frame and the second key frame.
[0004] "Collecting multiple frames of first images, identifying a first key frame from the multiple frames of first images, and there is a preset positional relationship between the target individual and the first identifier in the first key frame" specifically means: collecting multiple frames of first images at a fixed point, obtaining the first image position of the first identifier; identifying a first key frame from the multiple frames of first images, so that there is a preset positional relationship between the target individual and the first image position in the first key frame.
[0005] "Collecting multiple frames of second images, identifying a second key frame from the multiple frames of second images, and there is a preset positional relationship between the target individual and the second identifier in the second key frame" specifically means: collecting multiple frames of second images at a fixed point, obtaining the second image position of the second identifier; identifying a second key frame from the multiple frames of second images, so that there is a preset positional relationship between the target individual and the second image position in the second key frame.
[0006] "Identifying the first key frame in which there is a preset positional relationship between the target individual and the first identifier from the first image" specifically means: identifying the target individual from the first image, identifying the first key frame from the first image, and there is a preset positional relationship between the first moving individual and the first identifier in the first key frame, where the first moving individual is identified as the target individual through tracking.
[0007] "Identifying the second key frame in which there is a preset positional relationship between the target individual and the second identifier from the second image" specifically means: identifying the target individual from the second image, identifying the second key frame from the second image, and there is a preset positional relationship between the second moving individual and the second identifier in the second key frame, where the second moving individual is identified as the target individual through tracking.
[0008] The movement trajectory of the target individual is on a circular track. If the first identifier is used as the starting line, the second identifier coincides with the finish line of a complete lap relative to the starting line. "The movement distance of the target individual between the first identifier and the second identifier analyzed according to the preset rule" is the distance traveled by making one or more laps around the circular track.
[0009] It further includes: collecting multiple frames of third images, identifying the third key frame from the multiple frames of third images, and there is a preset positional relationship between the target individual and the third identifier in the third key frame; evaluating the performance of the target individual's movement process between the first identifier and the third identifier according to the movement distance of the target individual analyzed according to the preset rule and the acquisition times of the first key frame and the third key frame.
[0010] It further includes: evaluating the performance of the target individual's movement process between the second identifier and the third identifier according to the movement distance of the target individual analyzed according to the preset rule and the acquisition times of the second key frame and the third key frame.
[0011] "Identifying the third key frame in which there is a preset positional relationship between the target individual and the third identifier from the third image" specifically means: identifying the target individual from the third image, identifying the third key frame from the third image, and there is a preset positional relationship between the third moving individual and the third identifier in the third key frame, where the third moving individual is identified as the target individual through tracking.
[0012] "Identifying the third key frame in which there is a preset positional relationship between the target individual and the third identifier from the third image" specifically means: collecting multiple frames of third images at a fixed point, obtaining the position of the third identifier in the third image; identifying the third key frame from the multiple frames of third images such that there is a preset positional relationship between the target individual and the position of the third image in the third key frame.
[0013] Further included are: obtaining a first track number corresponding to a target individual in the first key frame / the second key frame / the third key frame; obtaining a second track number corresponding to the target individual in the second key frame / the third key frame / the first key frame; comparing the first track number and the second track number to determine whether the target individual has switched lanes.
[0014] Further included is: comparing the acquisition time of the first key frame corresponding to the first identifier serving as the starting line with the starting command time to evaluate the reaction speed of the target individual or to evaluate whether the target individual has violated the rules.
[0015] The motion evaluation method proposed by the present invention has the following technical effects:
[0016] It is not necessary for a referee to participate in starting the command and recording the starting time at the starting point, which is beneficial to the autonomous training of individual athletes. Moreover, even if a false start occurs, the result can be corrected. In addition, it is not limited to the starting point and the ending point or the starting point, and motion data can also be given to individual athletes in a segmented manner as a reference. Detailed implementation manner
[0017] The present invention proposes a track and field event evaluation method based on machine vision, including:
[0018] S1. Collecting multiple first images, identifying a first key frame from the multiple first images, and there is a preset positional relationship between the target individual and the first identifier in the first key frame.
[0019] Wherein, "there is a preset positional relationship between the target individual and the first identifier" specifically means: the distance between the center point of the bottom edge of the target individual's human body frame and the first identifier is less than a preset threshold, which can be understood as the target individual reaching the position of the first identifier or exceeding the first identifier.
[0020] S1 is specifically:
[0021] Identifying the target individual from the first image, identifying the first key frame from the first image, and there is a preset positional relationship between the first moving individual and the first identifier in the first key frame, wherein the first moving individual is identified as the target individual through tracking.
[0022] The method of combining tracking and identification provides room for choice in face recognition of the target individual, rather than necessarily identifying the target individual in the first key frame. Because it is not certain whether the face in the first key frame is facing the camera, this method can greatly improve the recognition rate.
[0023] There are two implementation manners.
[0024] Manner 1
[0025] Specifically included are:
[0026] S11. Continuously collect the first images containing the first identifier, and continuously identify multiple frames of the first images until the target individual is identified in a certain frame.
[0027] S12. Start tracking the target individual based on the certain frame image in which the target individual is identified above, until the first key frame is identified, where there is a preset positional relationship between the first identifier in the first key frame and the target individual.
[0028] Method 2
[0029] Specifically include:
[0030] S11. Continuously collect the first images containing the first identifier, track one or more moving individuals appearing in the first images and the key frames corresponding to the moving individuals, where there is a preset positional relationship between the first identifier and the moving individuals in the key frames.
[0031] S12. Continuously identify the first images, and identify the target individual included in one or more moving individuals in a certain frame or certain frames. The key frame corresponding to the identified moving individual is the first key frame.
[0032] S1 is specifically:
[0033] Fixed-point collect multiple frames of the first images, obtain the first image position of the first identifier; identify the first key frame from multiple frames of the first images, so that there is a preset positional relationship between the target individual and the first image position in the first key frame.
[0034] Here, fixed-point collection can be understood as that the actual area corresponding to the multiple frames of the first images collected remains unchanged. That is to say, the actual position of the actual area corresponding to the first identifier in multiple frames of the first images will not change either. In this embodiment, the first image position is the image position where the first identifier does not change in the first image. It can be obtained by including the first identifier in one of the first images and then through manual calibration and storage for later call, or by making the first identifier appear in each frame and then obtaining it.
[0035] The first identifier can coincide with the bottom boundary of the multiple frames of the first images collected at a fixed point.
[0036] S2. Collect multiple frames of the second images, identify the second key frame from multiple frames of the second images, and there is a preset positional relationship between the target individual and the second identifier in the second key frame.
[0037] Among them, "there is a preset positional relationship between the target individual and the second identifier" specifically means: the distance between the center point of the bottom edge of the target individual's human body frame and the second identifier is less than a preset threshold, which can be understood as the target individual reaching the position of the second identifier or exceeding the second identifier.
[0038] Specifically, S2 is as follows: identify the target individual from the second image, and identify the second key frame from the second image. There is a preset positional relationship between the second moving individual and the second identifier in the second key frame, where the second moving individual is identified as the target individual through tracking.
[0039] The method of combining tracking and recognition provides options for face recognition of the target individual, rather than necessarily identifying the target individual in the second key frame. Because it is uncertain whether the face in the second key frame faces the camera, this method can greatly improve the recognition rate.
[0040] There are two independent implementation methods.
[0041] Method 1
[0042] Specifically, it includes:
[0043] S21. Continuously collect the second image containing the second identifier, and continuously recognize multiple frames of the second image until the target individual is recognized in a certain frame.
[0044] S22. Start tracking the target individual based on the certain frame image in which the target individual is recognized above until the second key frame is recognized. Among them, there is a preset positional relationship between the second identifier and the target individual in the second key frame.
[0045] Method 2
[0046] Specifically, it includes:
[0047] S21. Continuously collect the second image containing the second identifier, track one or more moving individuals appearing in the second image and the key frames corresponding to the moving individuals. There is a preset positional relationship between the second identifier and the moving individual in the key frame.
[0048] S22. Continuously recognize the second image, and identify the target individual included in one or more moving individuals in a certain frame. The key frame corresponding to the recognized moving individual is the second key frame.
[0049] Specifically, S2 is as follows:
[0050] "Collect multiple frames of the second image, identify the second key frame from the multiple frames of the second image, and there is a preset positional relationship between the target individual and the second identifier in the second key frame" is specifically: collect multiple frames of the second image at a fixed point, obtain the position of the second image of the second identifier; identify the second key frame from the multiple frames of the second image, so that there is a preset positional relationship between the target individual and the second image position in the second key frame.
[0051] The fixed-point acquisition here can be understood as that the actual area corresponding to the multiple frames of the second image remains unchanged. That is to say, the actual position of the actual area corresponding to the second identifier in the multiple frames of the second image will not change either. In this embodiment, the position of the second image is the image position where the second identifier will not change in the second image. It can be obtained by including the second identifier in one of the frames of the second image and then obtaining it through manual calibration, and then storing it for later use. It can also be obtained by making the second identifier appear in each frame.
[0052] The second identifier can coincide with the bottom boundary of the multiple frames of the second image acquired at the fixed point.
[0053] S3. Collect multiple frames of the third image, identify the third key frame from the multiple frames of the third image, and there is a preset positional relationship between the target individual and the third identifier in the third key frame.
[0054] Among them, "there is a preset positional relationship between the target individual and the second identifier" specifically means: the distance between the center point of the bottom edge of the human body frame of the target individual and the third identifier is less than the preset threshold, which can be understood as the target individual reaching or exceeding the position of the third identifier.
[0055] S3 is specifically as follows:
[0056] Identify the target individual from the third image, identify the third key frame from the third image, and there is a preset positional relationship between the third moving individual and the third identifier in the third key frame, where the third moving individual is recognized as the target individual through tracking.
[0057] By combining the methods of tracking and recognition, there is a choice for face recognition of the target individual, rather than necessarily identifying the target individual in the third key frame. Because it is not certain whether the face in the third key frame is facing the camera, this method can greatly improve the recognition rate.
[0058] There are two independent implementation methods.
[0059] Method 1
[0060] Specifically include:
[0061] S31. Continuously collect the third image containing the third identifier, and continuously recognize the multiple frames of the third image until the target individual is recognized in a certain frame.
[0062] S32. Start tracking the target individual based on the certain frame image where the target individual is recognized above until the third key frame is recognized, where there is a preset positional relationship between the third identifier and the target individual in the third key frame.
[0063] Method 2
[0064] Specifically include:
[0065] S31. Continuously collect third images containing a third identifier, track one or more moving individuals appearing in the third images and the key frames corresponding to the moving individuals, where there is a preset positional relationship between the third identifier and the moving individuals in the key frames.
[0066] S32. Continuously identify the third images, and identify the target individual included in one or more moving individuals in a certain frame. The key frame corresponding to the identified moving individual is the third key frame.
[0067] Specifically, S3 is as follows:
[0068] Fixed-point collect multiple frames of third images to obtain the third image position of the third identifier; identify the third key frame from the multiple frames of third images, so that there is a preset positional relationship between the target individual and the third image position in the third key frame.
[0069] Here, fixed-point collection can be understood as that the actual areas corresponding to the multiple frames of third images collected are unchanged. That is to say, the actual position of the area corresponding to the third identifier in the multiple frames of third images will not change either. In this embodiment, the third image position is the image position where the third identifier does not change in the third image, which can be obtained by including the third identifier in one of the third images and then obtaining it through manual calibration and storing it for later use, or by making the third identifier appear in each frame and then obtaining it.
[0070] The third identifier can coincide with the bottom boundary of the multiple frames of third images collected at a fixed point.
[0071] S4. Perform performance evaluation on the movement process of the target individual between the first identifier and the second identifier according to the movement distance of the target individual analyzed according to a preset rule and the acquisition times of the first key frame and the second key frame.
[0072] Perform performance evaluation on the movement process of the target individual between the first identifier and the third identifier according to the movement distance of the target individual analyzed according to a preset rule and the acquisition times of the first key frame and the third key frame.
[0073] Perform performance evaluation on the movement process of the target individual between the second identifier and the third identifier according to the movement distance of the target individual analyzed according to a preset rule and the acquisition times of the second key frame and the third key frame.
[0074] Among them, the actual positions of the first identifier, the second identifier, and the third identifier can be pre-stored. Based on this, the movement distance of the target individual between any two of the first identifier, the second identifier, and the third identifier can be analyzed according to a preset rule. And based on the acquisition times of the first key frame, the second key frame, and the third key frame, the movement time of the target individual between any two of them can be associated; thereby evaluating the performance of the target individual in this process.
[0075] In S4, the movement trajectory of the target individual can be on a circular track. If the first identifier is used as the starting line, the second identifier can coincide with the finish line that completes a full circle relative to the starting line. "The movement distance of the target individual between the first identifier and the second identifier analyzed according to the preset rule" is the distance traveled by running one or more laps around the circular track.
[0076] Taking the example of a 1000-meter run on a 400-meter standard circular track, in the prior art, the result can only be calculated when reaching 1000 meters. However, in this embodiment, the target individual starts from the first identifier as the starting line. Whether it passes one lap of 400 meters, two laps of 800 meters, or reaches the finish line of 1000 meters, the result can be calculated, and the movement state of the target individual at each stage during the movement process can be evaluated for reference.
[0077] Taking the example of running multiple laps on a 400-meter standard circular track, in the prior art, the result can only be calculated when reaching the finish line. However, in this embodiment, the target individual starts from the starting line. Whether it completes one lap, two laps, etc. or finally reaches the finish line, the result can be calculated to realize the evaluation for reference of the movement state of the target individual at each stage during the movement process.
[0078] Based on the above description, it is not difficult to conclude that for any two identifiers, whether adjacent or not, in Embodiment 1, the performance can be evaluated. Of course, it can also be extended to scenarios with more than three identifiers and circular track scenarios.
[0079] S5. Obtain the first track number corresponding to the target individual in the first key frame / the second key frame / the third key frame; obtain the second track number corresponding to the target individual in the second key frame / the third key frame / the first key frame; compare the first track number and the second track number to determine whether the target individual has changed lanes.
[0080] S6. Compare the acquisition time of the first key frame corresponding to the first identifier used as the starting line with the starting command time to evaluate the reaction speed of the target individual or to evaluate whether the target individual has violated the rules.
[0081] The present invention is conducive to the autonomous training of the sports individual, and even if a false start occurs, the results can be corrected. In addition, it is not limited to the starting point, the ending point or the starting line, and the sports data can also be given in segments to the sports individual as a reference.
[0082] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.
Claims
1. A track and field evaluation method based on machine vision, characterized in that Including: Collecting multiple first images, identifying a first key frame from the multiple first images, and there is a preset positional relationship between the target individual and the first identifier in the first key frame; Collecting multiple second images, identifying a second key frame from the multiple second images, and there is a preset positional relationship between the target individual and the second identifier in the second key frame; evaluating the performance of the target individual's movement process between the first identifier and the second identifier according to the movement distance of the target individual analyzed according to a preset rule and the acquisition times of the first key frame and the second key frame.
2. The track and field event evaluation method according to claim 1, wherein "Collecting multiple first images, identifying a first key frame from the multiple first images, and there is a preset positional relationship between the target individual and the first identifier in the first key frame" specifically means: collecting multiple first images at a fixed point, obtaining the first image position of the first identifier; identifying a first key frame from the multiple first images, so that there is a preset positional relationship between the target individual and the first image position in the first key frame.
3. The track and field event evaluation method according to claim 1, characterized in that "Collecting multiple second images, identifying a second key frame from the multiple second images, and there is a preset positional relationship between the target individual and the second identifier in the second key frame" specifically means: collecting multiple second images at a fixed point, obtaining the second image position of the second identifier; identifying a second key frame from the multiple second images, so that there is a preset positional relationship between the target individual and the second image position in the second key frame.
4. The evaluation method according to claim 1, characterized in that, "Identifying a first key frame in which there is a preset positional relationship between the target individual and the first identifier from the first image" specifically means: identifying the target individual from the first image, identifying the first key frame from the first image, and there is a preset positional relationship between the first moving individual and the first identifier in the first key frame, where the first moving individual is identified as the target individual through tracking.
5. The track and field event evaluation method according to claim 1, wherein "Identifying a second key frame in which there is a preset positional relationship between the target individual and the second identifier from the second image" specifically means: identifying the target individual from the second image, identifying the second key frame from the second image, and there is a preset positional relationship between the second moving individual and the second identifier in the second key frame, where the second moving individual is identified as the target individual through tracking.
6. The track and field event evaluation method according to claim 1, wherein The movement trajectory of the target individual is on a circular track. If the first identifier is used as the starting line, the second identifier coincides with the finish line that completes a full circle relative to the starting line. "The movement distance of the target individual analyzed according to a preset rule between the first identifier and the second identifier" is the distance traveled by running one or more laps around the circular track.
7. The track event evaluation method according to claim 1, wherein Also including: Collecting multiple third images, identifying a third key frame from the multiple third images, and there is a preset positional relationship between the target individual and the third identifier in the third key frame; evaluating the performance of the target individual's movement process between the first identifier and the third identifier according to the movement distance of the target individual analyzed according to a preset rule and the acquisition times of the first key frame and the third key frame.
8. The track event evaluation method according to claim 7, wherein Also including: Evaluating the performance of the target individual's movement process between the second identifier and the third identifier according to the movement distance of the target individual analyzed according to a preset rule and the acquisition times of the second key frame and the third key frame.
9. The track event evaluation method according to claim 8, wherein "Identifying the third key frame in which there is a preset positional relationship between the target individual and the third identifier from the third image" specifically means: identifying the target individual from the third image, identifying the third key frame from the third image, and there is a preset positional relationship between the third moving individual and the third identifier in the third key frame, where the third moving individual is identified as the target individual through tracking.
10. The track and field evaluation method according to claim 7, wherein "Collecting multiple frames of the third image, identifying the third key frame from the multiple frames of the third image, and there is a preset positional relationship between the target individual and the third identifier in the third key frame" specifically means: "Collecting multiple frames of the third image, identifying the third key frame from the multiple frames of the third image, and there is a preset positional relationship between the target individual and the third identifier in the third key frame" specifically means: collecting multiple frames of the third image at a fixed point, obtaining the position of the third image of the third identifier; identifying the third key frame from the multiple frames of the third image, such that there is a preset positional relationship between the target individual and the third image position in the third key frame.
11. The evaluation method according to claim 7, wherein It further includes: Obtaining the first runway number corresponding to the target individual in the first key frame / the second key frame / the third key frame; Obtaining the second runway number corresponding to the target individual in the second key frame / the third key frame / the first key frame; comparing the first runway number and the second runway number to determine whether the target individual has switched lanes.
12. The evaluation method according to claim 7, wherein It further includes: Comparing the acquisition time of the first key frame corresponding to the first identifier serving as the starting line with the starting command time to evaluate the reaction speed of the target individual or to evaluate whether the target individual has violated the rules.