Sports Video Analysis Algorithm Based on Image Multiple Processing Technology
Jianyu Zhang · 2023
Considering the explosive growth in the amount of online video information in the Internet information era, and the fact that human actions in video are the main body of video information, there is a strong demand for mining in-depth information in video content. The application of multiple processing techniques for sports video images in sports research has further developed. This study builds a spatiotemporal dual attention network (ST-DAtt) recognition architecture based on the CNN-LSTM model and combined with the attention mechanism. This method utilizes features from multiple visual perception levels on CNN as input to a temporal model, and it uses different LSTM units to mine spatiotemporal context information. By introducing two attention models, the spatiotemporal saliency of the extracted features is enhanced. Finally, after feature dimensionality reduction, it enters the fusion model to fuse the output features of different networks, aiming to jointly enable decision-making capabilities at multiple levels and improve model recognition performance. The experimental results on multiple datasets demonstrate the effectiveness of the research model, as well as its strong spatiotemporal representation ability, fast model efficiency, and strong robustness.