Pedestrian detection and tracking of a mobile robot with multiple 2D laser range scanners

Abstract

This paper proposes a method for detecting and tracking pedestrians, which is an important function for preventing the collision of autonomously moving mobile robots with a pedestrian. To observe the pedestrian, we use two types of laser range scanners: one scans a 2D plane (2D scanner) and the other scans a 3D space, such as a 3D scanner. We scan 3D space with a 2D scanner fixed on the turning mechanism, instead of a 3D scanner, because 3D scanners are more expensive than 2D scanners. The 3D space is divided into voxels to determine whether a pedestrian exists. This detection system is robust to occlusion because turning 3D scanner can detect a pedestrian based on the shape of the data points. However, the scanning span of the turning 3D scanner depends on the period of the turning mechanism; thus, we combine the detected pedestrian's position with the data scanned by the normal 2D scanner because it is possible for a 2D scanner to track objects moving in real time according to the positional transition of the data points. To investigate the efficiency of the proposed method for pedestrian detection, which is robust to occlusion and exhibits real-time tracking using reasonable scanning devices, we experiment with our prototype.