An Improved Gravitational Edge Detection Approach for Crowd Density Estimation

To estimate high dense crowd density, an improved gravitational edge detection approach is proposed. First, the background noises are removed from original images. Second, the gravitational edge detection method is applied to obtain the gray edge image, and based on the histogram, the binary edge image can be got automatically. Finally, the length of edge and the number of grid in the edge image are utilized for crowd density estimation. Compared to other estimation methods, the proposed approach is better at dealing with highly crowded objects, which contains much overlaps. Experimental results show, it averagely takes 1.9s to handle 1s original video.