Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 08, 2016 - June 11, 2016
Every year, the number of automobile accidents and casualties are decreasing in Japan; however, around 4,000 people are still killed in fatal accidents. Against this background, much attention has been paid to researches on pedestrian detection. On the other hand, few research efforts have focused on cyclist detection although the number of cyclist fatalities is as many as 600. This research proposes a pedestrian and cyclist detection method by LIDAR-camera fusion. We first detect pedestrian and cyclist candidates by extracting clusters in the point cloud data with a size filtering. We then classify the candidates using a shape-image combined feature vector. In the cyclist detection, we prepare two models corresponding to the side and the front/rear view to cope with a large appearance change. In addition, we implemented the proposed method on a real vehicle and tested it in real environments to show that the method can detect pedestrians and cyclists on-line.
