Motion Recognition Model of Sports Video Based on Feature Extraction Algorithm

With the rapid development of computer technology, network technology and multimedia technology, multimedia data is increasing exponentially. As an important part of video multimedia data, its structure is complex and its data volume is huge. Especially from a monocular camera, it is a very meaningful work to realize the correct recognition of human movements in any visual angle. Video action recognition usually refers to the process of identifying human action categories from a video sequence. This technology is widely used in multimedia content analysis, human-computer interaction, intelligent real-time monitoring and other fields. It can be realized by feature extraction of video to generate feature vectors, and classification of feature vectors by classifier. To solve the above problems, this paper starts from the spatio-temporal correlation of video data, according to the mixed model structure of convolutional neural network (CNN) and long short term memory (LSTM), introduces the attention mechanism in semantic analysis, proposes a recursive machine network based on the attention mechanism, assigns different weights to each video frame, and improves the recognition effect in the captured video clips.