Moving Foreground Detection Approach Of Complex Scenes

Foreground detection is a significant step of information acquisition in intelligent surveillance.The task is to segment all the moving objects from complex scenes without any false targets andnoise interference. Due to non-stationary surveillance, foreground extraction becomes a complextask with challenge. The performance of foreground detection mainly depends on backgroundmodeling algorithm. In order to solve this problem, some background modeling approaches areproposed in this paper.First of all, an adaptive approach based on Gaussian mixture model is proposed. In thisapproach, the number of Gaussians can be controlled dynamically. And online EM algorithm isapplied to the method for estimating the parameters in Gaussian mixture model. At last, severalstrategies are proposed to control the learning rate of weight. Experimental results show that theforeground object can be detected effectively and rapidly, and the precision and recall ratio ofresults demonstrate superiority of the method to some related work.In order to increase precision and recall ratio, a non-parametric background modeling approachis proposed. This approach uses Parzen density estimation in non-stationary scenes, and theprobability density function is partially updated within the range of the window function based onthe observed pixel value. Experimental results show that the model adapts quickly to changes in thescene and foreground objects can be robustly detected.At last, a hybrid background model is proposed. This model consists of two different kinds ofbackground models. One is pixel-level background model which is robust for long-termillumination changes. The other is LBP background model which is robust for short-termillumination changes. The experimental results demonstrate superiority of proposed method to somerelated works.