Research On Key Algorithms In Tactical Intelligence Analysis System Of Karate Competition Video

The classification and identification of human behavior in digital video is a new and challenging research direction in the field of computer vision.Since its rise,the traditional research based on moving object feature extraction,motion template analysis and other technologies,in practical applications,are mostly seriously interfered by factors such as occlusion,light,scale,and differences between individual limb movements.These factors affect the accuracy of classification and identification.Recently,with the development of deep learning technology,more and more behavior classification and identification algorithms based on deep convolution network come out.Compared with the traditional methods,the deep network model has a greater improvement in performance.In sports,the tactical information analysis of competition video is becoming more and more important with the popularity of video acquisition equipment.In this paper,we use the behavior classification and identification algorithm based on the depth network model to analyze the corresponding tactical information of karate game video,which greatly improves the analysis efficiency and accuracy.The research of this paper is divided into two parts:(1)get the twodimensional or three-dimensional skeleton information of athletes from the karate game video;(2)use the skeleton data to build the convolution network model of graph to classify and identify the behavior.The main research work and innovation of this paper are as follows:(1)Dataset production.A large number of digital videos of karate athletes during confrontation training are collected,and the video clips of kicking,punching and moving in karate are cut manually.Then,making the dataset used by system through imitating the way of making Kinetics.(2)Three kinds of pose estimation algorithms are compared and studied,and the experimental analysis is used to judge.In this paper,Open Pose pose estimation algorithms will be laid the foundation of the accuracy of behavior classification and identification.(3)In this paper,a new algorithm of Delaunay triangulation is proposed to construct the convolution model of graph in depth network model,which makes full use of the ability of Delaunay triangulation to capture the information of spatial structure,and constructs a graph model with more feature information.The advantages of the model are analyzed through experiments.(4)Aiming at the actual game video,the paper tests and analyzes the research results of the paper and the application effect in several tactical intelligence analysis.By comparing features and model accuracy with other algorithms,the algorithm proposed in this paper has a better effect on behavior classification in this system.