Research on Behavior Recognition Based on Video-SwinTransformer

With the development of deep learning technology, video target behavior recognition has become an important research direction in computer vision, and video data not only contains rich spatial information, but also contains complex temporal dynamics, making how to effectively deal with spatiotemporal features in video a key challenge. In recent years, models based on 3D-Convolutional Neural Network and Vison Transformer have made remarkable achievements in the processing of video analysis tasks, but there are still challenges in the processing of long-term span and complex dynamic information. To this end, this paper introduces a hyperlabeling algorithm to enhance the local feature modeling, reduce the computational complexity and improve the perception of finegrained information.