Analysis of baseball behavior recognition model based on Dual-GCN improved by motion weights.

This research aims to address the poor performance in baseball behavior recognition, insufficient connection between characters, and low accuracy in baseball behavior recognition. A motion weight improvement model based on dual-graph convolutional network is proposed. The new model takes a dual-graph convolutional network for behavior recognition and key region segmentation of baseball video images, and enhances the correlation and contribution between characters through motion weights. The research results indicated that the new model performed best when the image frame rate was 1/2w, the width of the key area was 1/2H, and the number of key areas was 2. The highest accuracy was 94.84%, which was 12.06% higher than that of the hierarchical temporal depth model. After adding motion weights, the accuracy improved by 3.45%. The accuracy of baseball recognition using different models has been effectively improved. The new model can more effectively recognize baseball behavior, which has important guiding significance for behavior recognition in baseball sports.