The Research On Algorithms Of People Tracking

Visual tracking is currently one of the most active topics in computer vision. As U.N. attaching the importance to the social public safety, the safe intelligent detection system needs to be improved and become more and more effective in public. Then the research agency of many countries and some world famous companies are now paying much attention to this field. As an essential part of the human motion analysis system, the research in people detecting and tracking can provide broad potential use in the future and economic benefits to the society.People tracking include three important parts that are image preprocessing, people detection and moving object tracking. The main works in this paper are as follows:1. The original images that we get from the video are usually polluted by some noise because of some exterior or interior influence. So the images should be smoothed before we analyze them. In this paper, two new denoising methods are proposed based on some classical ones. The first one is to realize filtering the noise in the high frequency parts of the image by using wavelet transform and multistage median filter. The second method is based on double Haar wavelet and Lee shrinkage algorithm and a new idea about the choice of the window's direction is also proposed. Experimental results show that the proposed algorithms can get better denoising result especially for protecting the details of images.2. People detection is also a difficult part in video tracking which need to realize getting the object from the complicated background. The background is usually dynamic and static. We regard the freeway or indoor environment as static background because there are no other changes except the light. In the outdoor environment, complicated factors make the background dynamic such as the moving branch bowed by wind, the ocean wave and the cloud. Object detection in static background can realize by the method of background subtracting which just need update the original background. In dynamic background, we proposed a new method based on Bayesian model and the correlation of frames. The new method can separate the moving foreground and the dynamic background correctly and the results show that we can detection the clear object from the complicated dynamic background.After the people detection, we should analyze the movement of each object that is enclosed by rectangle. We define the center of the rectangle as the character of each people and calculate the position of next frame according to the coordinate in the current frame. Then we can estimate the contrail of the object.The paper is organized as following five parts. Firstly, Section 1 gives an introduction of the background, significance and development of the field of people tracking. And Section 2 introduces some classical methods in image processing and proposed two new ones. We also give some relative analysis and comparison in above methods. Then Section 3 offers knowledge about the object detecting. We respectively discuss the problems in static and dynamic background and give the new algorithm and the experiment result. Then section 4 gives the analysis of moving object and some experiment images. At last, the paper ends with some further research plans mentioned in Section 7.