LiDAR-Based Symmetrical Guidance for 3D Object Detection.

Object detection from 3D point clouds is an essential task for autonomous driving. Current approaches usually focus on the characteristics of the point cloud while ignoring the structural characteristics of the object itself. This paper designs a new symmetric structure enhancement network SS-PV-RCNN according to the symmetry of car structure, inspired by the human symmetry visual cognition system. Using the symmetrically strengthened point cloud to help the network learn the features of the symmetric structure, improve the robustness of the network in the occlusion area, and then improve the accuracy of 3D object detection. This method is verified on the KITTI dataset, and it has a particular improvement compared to the baseline method.