Title: 3D object detection based on image and LIDAR fusion for autonomous driving

Authors: Guoqiang Chen; Huailong Yi; Zhuangzhuang Mao

Addresses: School of Mechanical and Power Engineering, Henan Polytechnic University, Jiaozuo, Henan, China ' School of Mechanical and Power Engineering, Henan Polytechnic University, Jiaozuo, Henan, China ' School of Mechanical and Power Engineering, Henan Polytechnic University, Jiaozuo, Henan, China

Abstract: 3D object detection is the fundamental task of autonomous driving. The existing approaches are very expensive in computation due to high dimensionality of point clouds. We use the 3D data more efficiently by representing the scene from the RGB image and the Bird's Eye View (BEV). The whole network is composed of two parts: one is the 2D Proposal Network for 2D region proposal generation, and the other is the 3D Region-based Fusion Network to predict the 2D locations, orientations, and 3D locations of the object. First, we fuse the BEV feature map and the RGB image to enhance the input. Second, we adopt the 3D encoding form with 2D-3D bounding box consistency constraints and design ROI-wise feature fusion to predict location information. Our experimental evaluation on both the KITTI as well as a large scale 3D object detection benchmark shows signicant improvements over the state of the art.

Keywords: 3D object detection; image and LIDAR; deep learning; multi-sensor fusion; autonomous driving.

DOI: 10.1504/IJVICS.2023.132927

International Journal of Vehicle Information and Communication Systems, 2023 Vol.8 No.3, pp.237 - 251

Received: 12 Feb 2020
Accepted: 31 Jan 2021
Published online: 17 Aug 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article