Multi-directional crowded objects segmentation based on optical flow histogram

This paper proposes an adaptive crowded objects segment algorithm. First, a real time global optical flow method is employed to calculate velocity field and the background noise is eliminated by setting threshold. Second, the angle matrix of foreground optical flow is treated as gray scale image, and its histogram curve is segmented instead of optical flow foreground. Through the histogram derivative curve, a set of segment points is picked up, and then foreground area is segmented into different flows. During the segmentation, some small blocks appear. A block absorption approach is proposed to solve this problem, which makes use of the color characteristic of flows. Some experimental results show the algorithm is efficient. Compared to clustering methods, the proposed approach has similar segment result, but is very time saving. The approach is more suitable for real time applications.