Camera and LiDAR Fusion for Robust 3D Person Detection in Indoor Environments

In order to guarantee a harmonious coexistence of robots and humans, it is vital for the former to be able to navigate the environment in a safe manner, taking into consideration the people that populate it. This requires the robot to be capable of identifying where the surrounding people are, through adequate processing of the sensory data acquired. In this paper we present a 3D person detection system for indoor settings which fuses RGB and LiDAR data, simultaneously using the semantic information available from the camera and the precise and long-range data acquired by the LiDAR. The proposed approach utilizes SVM with various feature vectors to classify the point clusters, and it also introduces planarity as another feature, improving the overall detection performance considerably.