Abstract: Traditional pedestrian detection system uses global template for search and recognition, resulting in large useless search and reducing the realtime pedestrian detection performance. In addition, it is difficult to apply to unmanned vehicle and pedestrian safety technology in the traffic environment. Based on the difference of motion between pedestrian and vehicle moving background, a fat pedestrian area estimation method is proposed using optical flow clustering. This method uses the combination of local and global optical flow to obtain the optical flow field. According to the difference between background and foreground, the optical flow clustering algorithm and background estimation method are established. Then, the image segmentation algorithm is used to realize the segmentation of optical flow vector. Finally, according to the characteristics of the human body, we can judge whether the candidate region of the moving object is an effective area for pedestrians and realize the pedestrian area estimation. Experiments are carried out based on JLU-PDS and Daimler international pedestrian sharing test library, and experimental environments include the crowed, shelter, streets, fields and other auxiliary working conditions. Results show that the algorithm has good detection effect and greatly reduces the sear space of pedestrian recognition. In addition, it provides technical support and basis for research and application of vehicle active pedestrian protection technology and unmanned intelligent vehicles.

Key words: engineering of communication and transportation system, pedestrian detection, autonomous vehicle, smart vehicle, optical flow segmentation