Research And Implementation Of Improved Mean Shift Tracking Algorithm

Object tracking algorithms aim to find the most similar region with the target from an image sequence. Object tracking plays an important role for many applications like visual surveillance, artificial intelligence, human-computer interfaces and etc. It could provide reliable information for object action analysis by getting the track. In this thesis we focus on the Mean Shift algorithm in the object tracking field.Traditional Mean Shift is based on the assumption that the target object would not change a lot during tracking. It uses the initial template from beginning to end, without any template updating mechanism.This thesis adopts moving object detection method, which uses Mixture Guass Background Modeling, to update target template. This method makes template updated adaptively. Although the scale of the Mean Shift kernel is a crucial parameter, there is presently no mechanism for choosing or updating scale while tracking objects that changing in size. To solve this problem, the thesis uses variational level set method to get the contour of object and realizes the self-adaption of kernel scale according to the contour information. Based on the background of realizing intellectualized video surveillance system, this thesis studies and realizes algorithm for the No Parking area surveillance system based on the improved Mean Shift, mainly research on two key technologies in the intelligent video surveillancesystem, namely moving target detection and tracking.The following work has been done:First, studying Mean Shift theory and its application of tracking algorithm and raising problems in this algorithm. Secondly, for the two limitations in traditional algorithm, proposing improved algorithms. Then, studying and implementing the surveillance algorithm of No Parking area based on the improved Mean Shift.