The Construction of Mathematical Model of Swimmers’ Technical Movements using Multimodal Deep Learning Framework

Abstract

This paper proposes a mathematical model construction method based on a multi-modal deep learning framework aiming at the accuracy and real-time requirements of swimmers’ technical movement analysis. The model can extract the image features and timing information of athletes’ movements from video sequences by integrating spatiotemporal modules. This paper introduces the translation partial channel strategy to overcome the limitation of spatiotemporal information separation in traditional methods, which can seamlessly integrate spatiotemporal features and enhance the recognition ability of complex action patterns. In addition, NetVLAD is used as the feature aggregation layer. This layer can capture and encode the global and local features of the athlete’s movements, thereby improving the classifier’s performance. In the experimental part, the model is strictly verified, and the results show that compared with the prior art, the model in this paper shows higher accuracy and faster processing speed in the swimmer’s action classification task. This provides the possibility of immediate feedback for coaches and athletes and lays a solid foundation for further research in the field of sports science.