Search

CN-121999236-A - Shuttlecock round segmentation method, device, equipment and medium

CN121999236ACN 121999236 ACN121999236 ACN 121999236ACN-121999236-A

Abstract

The invention relates to the technical field of computer vision and discloses a shuttlecock round segmentation method, device, equipment and medium, which comprises the steps of obtaining video streams of a shuttlecock training field; the method comprises the steps of obtaining a detection result of shuttlecocks in each frame of image, carrying out object detection on the video stream to obtain the detection result of the shuttlecocks in each frame of image, wherein the detection result comprises the positions of the shuttlecocks and the motion state of the shuttlecocks, and the motion state comprises a flight state and a static state.

Inventors

  • WANG BAIRUN

Assignees

  • 福州善为智行科技有限公司

Dates

Publication Date
20260508
Application Date
20251222

Claims (10)

  1. 1. A shuttlecock round cutting method, characterized in that the method comprises the following steps: acquiring a video stream of a badminton training field; Object detection is carried out on the video stream, so that a detection result of the shuttlecock in each frame of image is obtained, wherein the detection result comprises the position of the shuttlecock and the movement state of the shuttlecock, and the movement state comprises a flight state and a static state; Performing field detection on the video stream to obtain a division result of a target area and a non-target area in an image; screening effective shuttlecocks which are positioned in the target area and in the flying state from each frame of image based on the detection result of the shuttlecocks and the division result of the area; And cutting out round video clips belonging to the round of the game from each frame of image based on the time information of all the effective shuttlecocks.
  2. 2. The method of claim 1, wherein the performing object detection on the video stream to obtain a detection result of the shuttlecock in each frame of image comprises: Splicing two frames of images which are continuous in time in the video stream to form a combined image; and inputting the combined image into a target detection model to obtain the detection result.
  3. 3. The method of claim 2, wherein the target detection model is created based on YOLOv architecture, the inputting the combined image to the target detection model, resulting in the detection result, comprises: Inputting the combined image into a feature extraction network of a model to perform feature extraction to obtain a plurality of feature images with different resolutions; And respectively inputting each characteristic diagram to a prediction head corresponding to each other to obtain the detection result output by each prediction head, wherein each prediction head comprises a prediction category branch, a prediction coordinate branch and a prediction motion state branch, and the prediction motion state branch is used for detecting the motion state of the shuttlecock.
  4. 4. The method according to claim 1, wherein performing field detection on the video stream to obtain a result of dividing a target area and a non-target area in an image, comprises: performing key point identification on a current frame image in the video stream, wherein the key points are points for defining the ground outline of the local site; converting the key points from a world coordinate system to a pixel coordinate system, and calculating a field edge line for representing the ground contour of the pair of local fields based on the pixel coordinates of the key points; Translating the field edge line to the external direction of the opposite field by a preset distance to obtain a non-field edge line; And determining the non-target area according to the space area cut by the non-field edge line in the current frame image, and taking the rest area except the non-target area in the current frame image as the target area.
  5. 5. The method of claim 4, wherein the predetermined distance is determined by: determining a true distance between the ground profile of the playing field and the ground profile of the non-playing field; And converting the true distance from a world coordinate system to a pixel coordinate system to obtain the preset distance.
  6. 6. The method according to claim 1, wherein the step of segmenting the round video clips belonging to the round from each frame image based on the time information of the appearance of all the valid shuttlecocks comprises: Aggregating the effective timestamps into a plurality of candidate time clusters by temporal proximity based on the effective timestamps of all the effective shuttlecocks, wherein adjacent timestamp intervals within each candidate time cluster are smaller than a first threshold; checking the validity of each candidate time cluster; determining a time range corresponding to each valid time cluster passing through verification as a checking time period of a checking round; And cutting out round video clips belonging to the round of the game from each frame of image based on the round time period.
  7. 7. The method of claim 6, wherein verifying the validity of the respective candidate time clusters comprises: calculating the discrete degree index of the effective time stamp in each candidate time cluster; And determining a candidate set with the discrete degree index smaller than a second threshold as the valid time cluster.
  8. 8. A shuttlecock round cutting device, the device comprising: The video stream acquisition module is used for acquiring video streams of badminton training sites; the object detection module is used for detecting the object of the video stream to obtain a detection result of the shuttlecock in each frame of image, wherein the detection result comprises the position of the shuttlecock and the movement state of the shuttlecock, and the movement state comprises a flight state and a static state; the field dividing module is used for carrying out field detection on the video stream to obtain a dividing result of a target area and a non-target area in the image; The effective badminton screening module is used for screening effective shuttlecocks which are positioned in the target area and in the flying state from each frame of image based on the detection result of the shuttlecocks and the division result of the area; And the round and segmentation module is used for segmenting round video clips belonging to the round of the game from each frame of image based on the time information of all the effective shuttlecocks.
  9. 9. An electronic device, comprising: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory having stored therein computer instructions, the processor executing the computer instructions to perform the shuttlecock round segmentation method of any of claims 1 to 7.
  10. 10. A computer-readable storage medium having stored thereon computer instructions for causing a computer to perform the shuttlecock round segmentation method of any of claims 1 to 7.

Description

Shuttlecock round segmentation method, device, equipment and medium Technical Field The invention relates to the technical field of computer vision, in particular to a shuttlecock round segmentation method, device, equipment and medium. Background In the scenes of badminton training data analysis, competition round intelligent statistics, competition auxiliary penalty and the like, accurate and rapid judgment of whether a game is in a running state is a core premise. The shuttlecock game round starts with the service and finally the ball falls to the ground. The related technology adopts a visual model to directly extract and analyze image characteristics, so as to realize the classification judgment of whether each frame of picture in the video is in a contrast state, and the output result is yes (in contrast) or no (not in contrast), and the accuracy of the result of the method is poor. Disclosure of Invention The invention provides a shuttlecock round segmentation method, device, equipment and medium, which are used for solving the problems that in the prior art, automatic judgment of a round state is inaccurate, the adjacent field interference resistance is poor and manual operation is relied on under a complex training scene. The method comprises the steps of obtaining video streams of a badminton training field, detecting objects on the video streams to obtain detection results of shuttlecocks in each frame of images, wherein the detection results comprise positions of the shuttlecocks and motion states of the shuttlecocks, the motion states comprise flying states and static states, detecting the fields of the video streams to obtain division results of target areas and non-target areas in the images, screening out effective shuttlecocks which are located in the target areas and are in the flying states from each frame of images based on the detection results of the shuttlecocks and the division results of the areas, and cutting out round video segments belonging to the round from each frame of images based on time information of occurrence of all the effective shuttlecocks. In an optional implementation manner, object detection is performed on a video stream to obtain a detection result of the shuttlecock in each frame of image, and the method comprises the steps of splicing two frames of images which are continuous in time in the video stream to form a combined image, and inputting the combined image into a target detection model to obtain the detection result. In an alternative implementation mode, a target detection model is created based on YOLOv architecture, a combined image is input to the target detection model to obtain a detection result, the method comprises the steps of inputting the combined image to a feature extraction network of the model to perform feature extraction to obtain a plurality of feature images with different resolutions, and inputting the feature images to a prediction head in one-to-one correspondence to obtain a detection result output by each prediction head, wherein the prediction head comprises a prediction category branch, a prediction coordinate branch and a prediction motion state branch, and the prediction motion state branch is used for detecting the motion state of the shuttlecock. In an alternative implementation mode, field detection is carried out on a video stream to obtain a division result of a target area and a non-target area in an image, the method comprises the steps of carrying out key point identification on a current frame image in the video stream, wherein key points are points used for defining a local field ground outline, converting the key points from a world coordinate system to a pixel coordinate system, calculating a field edge line used for representing the local field ground outline based on the pixel coordinates of the key points, translating the field edge line to the external direction of the local field by a preset distance to obtain a non-field edge line, determining the non-target area according to a space area cut by the non-field edge line in the current frame image, and taking the rest area except the target area in the current frame image as the target area. In an alternative embodiment, the predetermined distance is determined by determining the true distance between the ground surface profile of the arena and the ground surface profile of the non-arena, and converting the true distance from the world coordinate system to the pixel coordinate system to obtain the predetermined distance. In an alternative implementation mode, the method for segmenting the round video clips belonging to the round from each frame image based on the time information of all the valid shuttlecocks comprises the steps of aggregating the valid time stamps into a plurality of candidate time clusters based on the valid time stamps of all the valid shuttlecocks, wherein the adjacent time stamp interval in each candidate time cluster is smaller than a fir