RGB Based Crowded Pedestrian Counting in Red Light Time
G. Y. Zhang, Qijun Zhao, Zhipeng Luo, Hongbin Wei · 2009
During the red-light time, when pedestrians walk to the zebra line, the presence between pedestrian and vehicle is one of the main problems for pedestrian counting. Common approaches including edge detection, histogram of gradient, and segmentation based on histogram of intensity gradient are not feasible. After the review of the general pedestrian counting methods, a new algorithm for pedestrian counting from the crowd RGB picture is proposed in this paper. Firstly, using the sequential frames with differential color threshold as based templates of human head and shoulder detection for preliminary counting db3 wavelet is used for retrieving the features. and K-mean cluster is used for the centroid detection of feature points clusters. Additionally, the track of centroid is used for the filtering of pedestrians not walking towards the camera. The algorithm is applied in pedestrian alerting system of Eshan Road and Dongfang road in Shanghai. In 1000 alerting events, the error is below 22% and the related analysis result is presented.