Crowd counting estimation in video surveillance based on linear regression function

Nowadays the estimation of crowd density and people counting is a great focus under public security affairs in surveillance. A scheme for crowd counting estimation based on linear regression function is proposed in this paper. In the proposed scheme, the Light Effect Suppression Model (LESM), which effectively reduces the sensitivity to illumination change, is applied to extract the foreground. Besides the number of foreground pixels, a novel feature called Oriented Inner Edges (OIEs) is proposed in this paper to deal with the problem of occlusion. Moreover, a new method of perspective normalization is applied during the feature extraction procedure to reduce the deviation caused by perspective distortion to the greatest extent. The final experimental results show that the proposed scheme is effective and feasible.