Vision-Based Crowd Pedestrian Detection

This paper proposes a crowd pedestrian detection based on monocular vision. To handle with the challenges faced in crowded scenes, such as occlusion, this study combines multiple cues to detect individuals in the observed image. Based on the assumptions that the human head is generally visible and background scene is stationary, all circular regions in the segmented foreground mask are firstly extracted by an algorithm called circle Hough transform (CHT). Each circle is then considered as the head candidate and further verified whether it is exactly an individual or a false one by combining multiple cues. Matching a candidate to a several constructed pedestrian templates is firstly applied for verification. Then, two proposed cues called head foreground contrast (HFC) and block color relation (BCR) are incorporated for further verification. In the experiment, three videos are used to validate the proposed method and the results show that the proposed one lowers the false positives at the expense of little detection rate.