Abstract: Aiming at the problem that traditional background models take too much time on initialization and exist Ghost area, we propose a kind of new model based on pixel sampling and the sample consensus theory. During the initialization phase, we sample the historical pixel sequence many times to build the background model. In the model updating phase, we update both of the historic pixel model and the ViBe background models with the improved ViBe algorithm. Finally, we use sample consensus theory to calculate the pixel between the two background models and find the target. The contrast test confirms that, the algorithm works well in Ghost area mitigation and achieves low speed target detection effects when foreground displayed in the initialization phase.

Key words: background modeling, sample consensus, target detection