Research And System Implementation Of Human Behavior Recognition For Sports Video

With the development of society and science and technology,video has become an important medium for information dissemination in today’s society,and the intelligent analysis of video content has become a hot topic in the field of scientific research.Human action behavior recognition is one of the important research contents in video content understanding,and due to the certainty and normativeness of sports technical movements,the finegrained classification of sports movements and the fast and complex movements of athletes make its recognition more challenging,so technical action recognition in sports videos has received more attention.In current sports field,sports empowered by artificial intelligence has become a major development trend,and intelligent analysis of sports based on video data has great potential for development.This thesis focuses on sports scenes,first conducts in-depth research on human behavior recognition tasks based on video data,and then designs and implements an intelligent sports analysis system based on algorithm research.In the algorithm research of human action recognition,in view of the challenges of large changes in human targets and complex technical actions in sports scenes,this thesis proposes a spatio-temporal action detection method based on a multi-scale separated spatio-temporal attention mechanism.This method gradually extracts and fuses spatio-temporal multi-scale features in the encoder-decoder network architecture through the attention mechanism,which can effectively cope with the challenge of action recognition in sports scenes,and can perform end-to-end training and inference.At the same time,the separation design of spatio-temporal attention in the time and space dimensions,as well as the multi-scale encoder-decoder structure,can effectively reduce the computational complexity of the attention mechanism in the spatio-temporal dimension.The experimental results on the public dataset also verify the effectiveness and superiority of the algorithm model proposed in this thesis.Based on the research basis of human behavior recognition algorithms in sports,this thesis designs and implements an intelligent sports analysis system,which comprehensively uses a variety of deep learning algorithms for intelligent analysis of sports videos.The system has designed a web interface on the front end to visually display video analysis results,and the back end is responsible for logical functions such as request response,task distribution,algorithm scheduling,and data reading and writing.For the algorithm model and function,this system uses virtualization and microservice technology to serve them as services,and builds an algorithm service center,and realizes convenient algorithm service calls through service registration and service discovery.Through the flexible collocation of different algorithm function modules,this system can also be easily expanded to various sports scenes,as well as other fields such as intelligent sports classrooms,intelligent fitness and entertainment.