Font Size: a A A

Improved Mean Shift Target Tracking Method And Its Application

Posted on:2018-12-08Degree:MasterType:Thesis
Country:ChinaCandidate:J J SunFull Text:PDF
GTID:2348330542478634Subject:Computer technology
Abstract/Summary:PDF Full Text Request
It is an important application field for the intelligent video surveillance in computer vision,which is widely used in security monitoring in the public area,such as communities,supermarkets and banks.Intelligent video surveillances contain target detection,target tracking,target recognition and behavior analysis.In the target tracking,the complexity of the background,the diversity of the characteristics of the tracked target itself,and the mismatched template caused by occluding the tracked target affects the performance of the target tracking algorithm.On the basis of the experimental research about Mean Shift target tracking algorithm,TLD target tracking algorithm and Kalman filter algorithm applied to target tracking,we discuss change of movement targets' attitude and their shape,shadow in different spatial areas,partially or completely blocked,light and scene,and real-time tracking,and we design a new target tracking algorithm,namely the MTKal algorithm.It aims to effectively improve the accuracy of target tracking,target recognition rate and the real-time performance,and reduce the time complexity.We study Mean Shift target tracking algorithm and the TLD target tracking algorithm respectively in experiments.Then we propose an improved Mean Shift target tracking algorithm,namely the MS-TLD algorithm.In order to verify the effectiveness of the MS-TLD algorithm,comparative experiments are designed in conditions of change of targets' scale and their simple background,interference in the target scene,and the fast moving target tracking respectively.Target recognition rate and tracking accuracy show that the performance of the MS-TLD algorithm is improved.We optimize the MS-TLD algorithm with the Kalman filter algorithm,and further propose the MTKal algorithm,in which the prediction function of the Kalman filter is used to predict targets' states by their existing motion information before iteratively computing the initial targets' candidate position in each frame on the Mean Shift target tracking algorithm.We compare with the experimental results of MS-TLD algorithm and the MTKal algorithm.Experiments show that MTKal algorithm can effectively improve the target tracking accuracy.The research result shows that the MTKal algorithm and the MS-TLD algorithm are used into target tracking systems.
Keywords/Search Tags:Moving Ttarget Tracking, MTKal Algorithm, MS-TLD Algorithm, Target Recognition Rate, Tracking Accuracy Rate
PDF Full Text Request
Related items