Swimming entry parameter detection method and system based on deep learning and projective transformation

Abstract

The application belongs to the technical field of water entry parameter detection, and discloses a swimming water entry parameter detection method and system based on deep learning and projection transformation, which comprises the following steps: determining the position coordinates of visual anchor points installed around a swimming lane in a swimming lane coordinate system, obtaining video frames of the water entry process of an athlete, and each video frame containing a visual anchor point; obtaining the pixel position coordinates of each limb joint of the athlete by using a deep learning algorithm based on the video frames; obtaining a projection matrix H from the image coordinate system to the swimming lane coordinate system based on the pixel position coordinates of the visual anchor points in the swimming lane coordinate system and the image coordinate system; judging the water entry time of the athlete based on the pixel position of the hand of the athlete in the video frames and the cross product, converting the pixel position coordinate point of the hand of the athlete at the water entry time into the position coordinates of the water entry time in the swimming lane coordinate system based on H, and completing the determination of the position parameters of the water entry of the athlete. The application can detect the swimming water entry parameters in real time and with high precision in a complex scene.

Images