Automatic recognition system of setting motion in volleyball setters and method thereof

Abstract

This invention provides an automatic method for recognizing volleyball ball-raising actions, comprising: capturing images using a camera; receiving the images using a processing device and processing them according to a skeleton detection algorithm to generate two-dimensional skeleton information; processing the two-dimensional skeleton information using a transformation algorithm to generate three-dimensional skeleton information; dividing the data sequence set into images using a segmentation method; and inputting the images into a learning model using the processing device to generate volleyball ball-raising action recognition results, and classifying the volleyball ball-raising actions as belonging to one of the volleyball ball-raising actions. Volleyball ball-raising actions include positioning actions, crouching positioning actions, ball-touching and releasing actions, and other actions. This method can effectively recognize volleyball ball-raising actions and is both real-time and objective.

Claims

1. An automatic recognition system for volleyball ball-raising actions, used to identify the actions of a player raising the ball, the automatic recognition system for volleyball ball-raising actions comprising: a plurality of cameras, positioned facing the player raising the ball, for capturing a plurality of images respectively; and a processing device connected to the cameras and receiving the images, the processing device processing the images according to a skeleton detection algorithm to generate a plurality of two-dimensional skeleton information, and processing the two-dimensional skeleton information according to a transformation algorithm to generate three-dimensional skeleton information, and dividing a data sequence set corresponding to the three-dimensional skeleton information through a segmentation method and converting it into an image, and inputting the image into a learning model to generate a volleyball ball-raising action recognition result, and classifying the volleyball ball-raising action recognition result as belonging to one of the plurality of volleyball ball-raising actions; wherein... The volleyball ball-raising action recognition result corresponds to the action of the player raising the ball. These volleyball ball-raising actions are a positioning action, a crouching positioning action, and a ball-touching and releasing action. The division method includes a sliding window method, which involves dividing the data sequence set into multiple windows and moving each window with a fixed step size. The data sequence set is a collection of all frames in time series. Each window contains multiple frames, each frame contains multiple keypoint coordinate information, the fixed step size is 8, the number of frames is 48, the number of keypoint coordinate information is 25, the image contains the keypoint coordinate information of one of the frames, the image is a heat map image, and each keypoint coordinate information contains three-axis information. The automatic recognition system for volleyball ball-raising actions further includes: A pressure sensing device is installed on the player lifting the ball to sense the pressure on the sole of the player's foot and generate a foot pressure signal; and a microcontroller is connected between the pressure sensing device and the processing device. The microcontroller receives and processes the foot pressure signal to generate digital pressure data and transmits the digital pressure data to the processing device; wherein, the processing device inputs the digital pressure data and the image into the learning model to generate the volleyball lifting action recognition result.

2. The automatic recognition system for volleyball ball-raising actions as described in claim 1, wherein the skeleton detection algorithm is OpenPose and the transformation algorithm is a Direct Linear Transformation (DLT) algorithm.

3. The automatic recognition system for volleyball ball-raising actions as described in claim 1, wherein, The learning model is a vision transformer. The ViT model; and the positioning action, the squatting and positioning action, and the ball contact and release action are different from each other.

4. An automatic method for recognizing volleyball raising actions, used to identify the actions of a player raising a ball, the automatic method for recognizing volleyball raising actions comprising: capturing a plurality of images using a plurality of cameras, wherein the cameras are positioned facing the player raising the ball; receiving the images using a processing device, and processing the images according to a skeleton detection algorithm to generate a plurality of two-dimensional skeleton information; processing the two-dimensional skeleton information using a transformation algorithm to generate three-dimensional skeleton information; dividing a data sequence set corresponding to the three-dimensional skeleton information using a segmentation method and converting it into an image; and inputting the image into a learning model using the processing device to generate a volleyball raising action recognition result, and classifying the volleyball raising action recognition result as belonging to one of the plurality of volleyball raising actions; wherein... The processing device is connected to the cameras. The volleyball ball-raising action recognition result corresponds to the action of the player raising the ball. These volleyball ball-raising actions are a positioning action, a squatting positioning action, and a ball-touching and releasing action. The division method includes a sliding window method, which involves dividing the data sequence set into multiple windows and moving each window with a fixed step size. The data sequence set is a collection of all frames in time. Each window contains multiple frames, and each frame contains multiple key point coordinate information. The fixed step size is equal to 8, the number of frames is equal to 48, the number of key point coordinate information is equal to 25, and the image contains the key point coordinate information of one of the frames. The image is a heat map image, and each key point coordinate information contains three-axis information. The automatic recognition method of the volleyball ball-raising action further includes: sensing the pressure on the sole of the player's foot using a pressure sensing device to generate a foot pressure signal. A microcontroller receives and processes the foot pressure signal to generate digital pressure data, and transmits the digital pressure data to the processing device; and the processing device inputs the digital pressure data and the image into the learning model to generate the volleyball lifting action recognition result; wherein, the pressure sensing device is installed on the lifting player, and the microcontroller is connected between the pressure sensing device and the processing device.

5. The automatic identification method for volleyball ball-raising actions as described in claim 4, wherein the skeleton detection algorithm is OpenPose and the transformation algorithm is a Direct Linear Transformation (DLT) algorithm.

6. The automatic recognition method for volleyball ball-raising actions as described in claim 4, wherein, The learning model is a Vision Transformer (ViT) model; and the positioning action, the squatting and positioning action, and the ball contact and release action are different from each other.

Images