Dynamic objects detection and tracking for a laser scanner and camera system

Abstract

Dynamic object detection and tracking from complex scenes is a challenge in the field of robot vision. In this paper, we exploit a laser scanner associated with a camera to achieve the map reconstruction, removing and tracking dynamic objects. Laser-based and image-based dynamic detection method are proposed, and their quality is analyzed respectively. Efficiency and accuracy are enhanced by a hybrid method associated 3D and 2D information. A 3D Gaussian mixture probability hypothesis density-based (GM-PHD) filter is designed to track the motion trajectories of multiple dynamic objects. The experimental result conducted on KITTI public dataset show the effectiveness of our method.