Max Intensity Voxelization Method for Autonomous Driving 3D Object Detection Based on LiDAR

In this paper, we propose a max intensity voxelization method for LiDAR-based 3D object detection in autonomous driving. First, it describes conventional voxelization methods and the proposed approach. Second, it explains PV-RCNN, one of the representative deep learning models for LiDAR-based 3D object detection. The experimental results indicate that the proposed voxelization method improves AP by approximately 0.31 compared to existing methods. It is expected that the proposed voxelization method will be suitable for most LiDAR-based 3D object detection deep learning models that require autonomous driving solutions.