Human body segmentation in complex situation is an important processing stage in video surveillance. The results have great affects on tracking, behavior recognition or object identification. In this paper, we present a method to segment region in which human bodies are overlapping or walking together. The multiple human objects region is segmented using potential function clustering in the vertical projection histograms of the region, the segmented results are verified by analyzing their geometry features. If the results are not consistent with the human geometry feature, it shows that multiple humans are overlapping. A binocular vision algorithm is proposed to segment the region based on distance. The Experimental results prove the algorithm works robustly, adaptively and efficiently.