Movement Detection And Object Tracking In Video Sequences

Research of digital image processing and computer vision is developing very rapidly in recent years. It is now widely used in military and civil application, and is becoming the most important way for intelligent machines to access external information and understand the world. For instance, the security supervision for important place, industrial processes' dynamic monitoring, dynamic robot vision and so on. Movement detection and object tracking are the two most important implications of computer vision, so there will be great practical significance and practical value on the research of them. It is also the research of this thesis.The selection of this paper is based on one of internal tasks that belong to Shandong computer science center .The name of it is: Intelligent Surveillance System. Our purpose is to find automatic, stable and efficient algorithm for the intelligent surveillance applications which can satisfy the robustness of the motion detection and the other more advanced applications.In this paper, we can describe researches as follows:For movement detection, We firstly propose a improved uni-Gaussian Model method which can extract complete information of moving objects, its computation is simple, and it is effective to the static and simple scene(like indoor).On the basis of it, the second mixed thresholding method is proposed in view of the dynamic and complex scene. Our method can obtain clear and complete information of the moving object, and eliminate noise fundamentally for the scene at the same time. Experiment results showed that motion detection with this thresholding method was accurate and real-time.For combine moving regions, the nearest distance of combine moving regions method was proposed. A moving object is always divided into several isolated regions in the process of movement detection, so combining moving regions is necessary, that is to say, two or more regions will be combined to one if they are near enough. In this article, we combine two or more regions to one when the nearest distance of them is below a threshold.For object tracking, an improved and efficient object tracking algorithm based on Camshift is proposed in the paper. For shortcomings of Camshift, we raise the tracking speed by virtue of predicting the position that a moving object arrives at the next time and reducing the search region. In the tracking algorithm, an acceleration equation is calculated for estimating the new position of a moving object, and a formula of predictive error is used to adjust the moving object search region automatically. In order to predict the future position accurately and simplify the computation, several motion parameters such as velocity and acceleration are updated adaptively by using IIR filters each frame. Several experiments are given to show that the proposed algorithm can improve object tracking speed by means of predicting a moving object position and reducing search region even if this moving object undergoes accelerated motion.