Football Tracking Algorithm Research In Soccer Video Based On Particle Filter

Soccer is one of the most popular sports in the world,and the research and analysis of soccer match video has been paid more and more attention by researchers and scholars.The research and analysis of soccer match video mainly use video image processing method to extract and analyze the target of interest in the video.In general,we regard soccer as tracking target,it is necessary to process video frame if we want to track the soccer,then we begin to track.So how to make the soccer tracking effect have high degree of robustness is a problem urgent needed to solve.According to the application and present situation of image processing method and the literature about the target detection tracking algorithm,we propose a soccer tracking algorithm based on particle filter.The main research contents are as follows:(1)Video frame pre-processing research.We transform the soccer match video into a sequence of video frames,then we make pre-processing operation on each frame image.Firstly,we converse the video frame from RGB space to HSV space;and then according to the characteristics which soccer usually appear in course rather than off-site and the main color of the course is mostly green,we extract the course and make binarization on it;and then through the Hough transformation,we detect the course and eliminate the impact on football detection of ball line;after morphological processing,the soccer detection pre-processing work is completed.(2)Research on rough selection algorithm.After pre-processing work on video frame,the course area retained only have players and football,we use the feature extraction algorithm,making roughing operation on soccer in video frame based on the shape,size and other characteristics of soccer.It can reduce the football tracking workload and improve the tracking efficiency.(3)Research on Particle Filtering Algorithm of soccer tracking.After the roughing,take the first frame sequence to extract the real football color characteristics as a template.Then infuse the feature template into the framework of particle filter algorithm,tracking every candidate soccer with the particle filter algorithm.Compare the sample characteristics with the templatefeatures,the higher the similarity,the greater the probability of real football.Then the highest similarity of the candidate ball will be tracked by a rectangular box.(4)Implementation and experiment of the algorithm.Through the experiment,the algorithm proposed in this paper and Mean-shift classical algorithm are compared,and in different light intensity,have a good feasibility.Especially in the case of football is blocked,the algorithm has a high robustness.The proposed algorithm has two innovations.The first innovation,roughing the work of football before track the football,greatly reducing the workload of football tracking.The second innovation,when applying particle filter to candidate soccer,this paper proposes an algorithm to limit the rectangle to jump,and improve the accuracy of football tracking.The shortcoming of the study is,in the case of football is blocked,the algorithm tracking effect is not very ideal.