R-CNN Network for Swim Athlete Action Recognition: A Deep Learning Method

Classification and object localization based on machine learning and computer vision technology have always been a hot research topic and have achieved great achievements. Professional athletes have never stopped exploring their pursuit of physical exercise and sports skills, and people of all ages are increasingly valuing sports. Researchers have responded to this need by integrating deep learning with the study of motion. The action recognition of swimmers in competitive swimming is the subject of this paper. This work offers an enhanced version of Mask R-CNN for the image-based job of swimmer action recognition. The model bilinear interpolates the extracted candidate regions using ROIAlign, recognizes and classes objects using segmentation branches, and finally employs an enhanced depth residual network (ResNet) as the feature extraction backbone network. The model enhances network performance by switching out the original standard convolution in the segmentation branch for deep separable convolution. In order to simulate algorithms, the experiment makes use of a network dataset. The experimental results demonstrate that the detection accuracy is improved by deepening the network and using deep separable networks while the network’s dependability is preserved. It proves that the enhanced Mask R-CNN network works as intended.