Research On The Algorithms Of Moving Object Detection And Tracking In Video Sequences

Moving target detection and tracking in video sequences has been widely concerned in the field of computer vision research hot topic,and widely applied in the precision guided weapons,smart surveillance,human computer interaction,intelligent robots and other military and daily life.This subject not only involves multiple disciplines,and its application in real is not the same.Although people have made a lot of research on this subject,but still there are some important problems unsolved.This thesis not only to the predecessors of the detection and tracking algorithm is carried on the thorough analysis and research,and in some ways their improved algorithm are given,to make it more perfect.In terms of moving target detection,this paper analyzes the three kinds of commonly used moving object detection method,the optical flow,interframe difference and background subtraction division.Of background subtraction based on gaussian mixture model division has made the analysis and research,this method mainly includes the foreground detection,post-treatment of pixel level and regional analysis and post-processing,feature extraction of five links.Through the analysis and the study found that the traditional background subtraction based on gaussian mixture model division,there are defects in the foreground detection link,slow convergence speed,and the algorithm requires a lot of storage space.Then on the basis of foreground detection algorithm for the concrete implementation steps of the algorithm is improved,and presents a new method of gaussian mixture model initialization,the method of using k-means clustering are online initialized on gaussian mixture model,at the same time the model updating method to do the further improvement and optimization,the matching criterion and new gaussian distribution to generate code to do the improvement.Through foreground detection simulation experiment found that the improved algorithm not only improves the convergence speed of the detection algorithm,and has good stability,in addition,from the algorithm on the runtime storage space occupied,experiments prove that saves almost half of the storage space.In terms of movement target tracking,in this paper for moving target tracking algorithm based on Kalman filter and using color histogram for tracking the characteristics of the Camshift tracking algorithm to do a thorough research and analysis.Through the analysis found that around when moving targets with similar color features of the large area of distractors,Camshift tracking algorithm can't accurate tracking moving targets.In order to solve this problem,this paper presents a Camshift tracking algorithm and tracking algorithm based on Kalman filter is the combination of new moving target tracking algorithm,through the experiment proves that the new algorithm can effectively solve the problem of large area color interference.