Adaptation Detection Algorithm In Complex Backgrounds Prospects Realization

With more and more people paying attention to the society safety, the visual surveillance system which are highly correlated with security problem, are also gained more attention. Foreground detection technique is one of the most essential tasks in a visual surveillance system. It is a prerequisite for object tracking, recognition, and localization, which enjoys widespread applications in visual surveillance, robotic navigation, military surveillance, traffic detections, and medical image analysis. The thesis primarily studies foreground detection algorithms in complex background, and proposes three algorithms to detect foreground in complex background.The traditional mixture Gaussian model which based on pixel, is often affected by noise and false positive, and could not deal with the abnormal movement and shadow of the foreground. Baseing on an algorithm which combines the region-based detection and pixel-based detection, the thesis proposes multiple reference models, which can effectively detect the foreground objects that have abnormal movements and can remove the ghost. And the thesis proposes an algorithm for removing shadowsThe thesis researches a foreground detection algorithm based on non-parameter density estimation, which could solve the selectition of the sample and the threshold problems in traditional algorithm. Firstly, the thesis improves a corner descriptor called SIFT-Displacement, and researches a sample selection algorithm and an adaptive threshold boundary algorithm. The thesis proposes a foreground detection algorithm which combined corner analysis with kernel density estimation.The thesis researches an algorithm which combined postposition filter technique and optical flow method. Firstly, the thesis proposes a predigested optical flow method, sloving the problem that the traditional algorithm needs clustering and highly cost. And the thesis proposes a postposition filter algorithm. The thesis implements a predigested optical flow method which based on postposition filter to detect foreground.