Research On Moving Object Tracking Based On Layered Templete Matching

The technique of moving object tracking(TMOT) is one of cores in the area of computer vision and it is significant and has broad applied value. This subject involves advanced techniques and scientific fruit of image processing, mode recognizability, human intelligence, automatic controlling and computer engineering. In essential, TMOT recognizes the object, abstracts its position information and tracks it automatically in image processing. Its difficulty lies in the information loss when the image signals are collected and when it happens in complicated environment. Especially, the problem of occlusion when signal is being collected affects the application very much, which becomes the emphasis of this paper.Tracking and measurement of moving object includes: capture (getting initial position), recognizability (or abstracts), position measurement and real-time tracking. Tracking and measurement supplement each other. To track the object, we must measure its position. However, tracking the object precisely is the precondition of precise measurement. Of course, capturing and recognizing object in time is the necessary point for stable tracking. So, when we discuss the arithmetic we should consider these aspects: initial position, abstracting features and matches, etc. These arithmetic can be separated into two parts: object recognizability and tracking, which can be not considered separately because the good recognizability arithmetic decides the stability of tracking, at the same time, the good tracking arithmetic can improve the working function of recognizability. Tracking algorithm mainly solve the problem of some object identification which this algorithm bases on, referring to feature analysis, tracking estimation and stability and so on. The paper discusses present popular algorithm, analyze and compare them and takes further consideration the method of model matching for object tracking. Layering Templete Match(LTM), namely we think the outer object as the center and separate many rectangle area and make sure that the outer has less weight value. This likes the pyramid which is cut even that the hiight represents weight, top is the central part and bottom is the outer. So center is considered to be much more important and less affection of the outer. This paper experience on PC .The images processed in this paper are from CCD camera which records airplanes and rockets. Because the condition is not good, noises affects the images in the process of getting, transforming and transition to form visual edges, so it is necessary to do pretreatment on images to improve all of these so that they are suitable for analysis by us or computer. Effective object features mainly depend on the pretreatment that influences the results'precise and processing speed. The paper filters, enhances , segments and does morphological processing on the original images to reduce noise.The emphasis discussed in paper is about occlusion. How to solve the problem is always a difficulty. It happens in some direction and scale new greyhound in the area of the tracking object images that reduce the proportion. This process goes gradually and effects tracking, even make the object lose. When the object is occlusioned partly, we can not get all the information so that there are two points for the algorithm of tracking: first, the algorithm must be good robust that can control affection from the loss to track the object ceaselessly. Second, the algorithm must process the problem of occlusion. However, there are so many kinds of occlusion and limited solutions of robust, if there is no answers for occlusion, we could not get our satisfying result.Layers Templete Matching is the center of algorithm, and surrounding it ,we propose a method based on LTM and fuzzy theory Kalman filter to process the sheltering. First, LTM is the basis and fuzzy theory will do update or reduction to judge occlusion and forecast the moving circuit using Kalman filter. We experience using MATLAB and the result shows that this method can solve the problem of occlusion, and real-time tracking when followed by false object. Compared to the method which adopts LTM only, this algorithm is much preciser. We research and discuss some key technique and algorithm from both theory and practice.