Extraction of individual pedestrians employing stereo camera images
Mitsuhisa Kawabe, Joo Kooi Tan, Hyoung Seop Kim, Seiji Ishikawa, Takashi Morie · International Conference on Control, Automation and Systems · 2011
A traffic accident occurs mainly because of carelessness of a driver. To prevent such an accident, many systems which detect pedestrians from vehicle are proposed. But most of these systems give a warning to a driver every time when a pedestrian is detected. The effect of warning becomes lower if the frequency of warning increases in a busy area. The objective of this study is to develop a technique for detecting pedestrians from a vehicle and to arrange them according to their potential risks by analyzing their behaviors. In this paper, we propose a technique for segmenting pedestrians employing stereo camera images to analyze a pedestrian's behavior. First, foreground regions are extracted employing background estimation using the mixture of Gaussian model. After extracting foreground regions in an input image, these regions are segmented by using the distance information. By using this method, each shape of the detected pedestrians can be obtained even when pedestrians are overlapping with each other. The proposed technique was examined experimentally employing real video images and satisfactory results were obtained.