An efficient 3D pedestrian detector with calibrated RGB camera and 3D LiDAR

Pedestrian detection plays an important role in the environmental perception and autonomous navigation for robotics, which provides critical information for the safe operation in complex environments. In this paper, a 3D pedestrian detector with calibrated LiDAR and RGB Camera is proposed, which takes full advantage of the precise range information of 3D LiDAR scanner and semantic information acquired from RGB image. The proposed approach integrates the segmented object clusters of point cloud and 2D bounding boxes generated by a visual object detector. The point cloud is segmented by a three-step segmentation approach, which can segment point cloud with both high precision and high efficiency. By fusion multi-sensor information in the image domain, the proposed approach provides 3D information of pedestrians in both LiDAR and camera coordinate systems. Experiments was conducted to evaluate the performance of the proposed approach.