Design And Implementation Of Sports Videos Analysis System Based On Deep Learning

In the sports,the standard of action and the energy conversion in the whole process of sports are of great concern.When assessing whether the movement of sports standards and energy conversion is reasonable in the sports,the real people's sports video should be transferred into 3D models in order for the relevant calculation for enengy conversion.With the popularity and development of sports,the traditional 3D model acquisition method of high-speed cameras with corresponding wearing equipment or 3D model reconstruction through depth-camera have been difficult to meet the current increasing audience in both efficiency and cost.At present,with the rapid development of the computer vision with deep learning technology,the use of deep learning technology to analyze sports videos provides more research value for 3D model reconstruction with high efficiency and low cost.This thesis studies and applies semantic segmentation algorithm and key node detection algorithm in the field of computer vision in deep learning.For fine-grained object segmentation,this thesis proposes the key algorithm SISA(Sports Image Segmentation Algorithm)which improves and optimizes the semantic seggmentation algorithm by using atrous convolution,pyramid pooling and trainable conditional random field structure.The key node detection algorithm is used to enhance the segmentation effect of small objects.Based on the above algorithm,this thesis designs and implements a sports video analysis system for extracting silhouette of sportsmen in the video.Users can do supervised training with annotation data to generate a deep learning model for extracting target semantics with system's training modules.The model can also be used to calculate and analyze the silhouettes of sportsmen in the video through the system's prediction module.This thesis first introduces the research background and research significance.This thesis studies and implements a variety of semantic segmentation techniques,and analyzes the advantages and disadvantages of each algorithm.And elaborates the implementation of SIS A used in this thesis and the optimization method.This thesis completed the requirements analysis and design of the whole sports video analysis system,designed the related test cases,and carried out the function and performance testing on the SIS A and the sports video analysis system.Finally,the development of sports video analysis system is summarized,and given the thinking and suggestions for the next step.