Video surveillance is of fundamental and critical importance for analysis and understanding interesting objects. An improved sampled-based model in this paper is introduced to detect these objects.