Hawk-Eye Deblurring and Pose Recognition in Tennis Matches Based on Improved GAN and HRNet Algorithms

Abstract

In tennis matches, the Hawk-eye system causes blurry trajectory judgment and low accuracy in player posture recognition due to rapid movement and complex backgrounds. Therefore, the research improves the backbone network and iterative attention feature fusion mechanism of deblur generative adversarial network version. At the same time, Ghost, Sandglass module, and coordinate attention mechanism are used to optimize the high-resolution network, and a new model for deblurring and pose recognition of Hawk-eye images in tennis matches is proposed by integrating the improved generative adversarial network and high-resolution network. The new model achieved an information entropy value of 11.2, a peak signal-to-noise ratio of 29.74 decibels, a structural similarity of 0.89, a minimum parameter size of 4.53, and a running time of 0.25 seconds on the tennis tracking dataset and the Max Planck Society human posture dataset, which was superior to current advanced models. The highest accuracy of deblurring and pose recognition for the model under different lighting intensities was 92.44%, and the highest improvement rate of video frame quality was 18%. From this, the model has significant advantages in deblurring effect, posture recognition accuracy, parameter quantity, and running time, and has high practical application potential. It can provide an advanced theoretical reference for tennis match refereeing and technical training.