3D object detection from point cloud

• Main Author: Yin, Xiangyu
• Other Authors: Jiang Xudong
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering::Computing methodologies
• Online Access: https://hdl.handle.net/10356/161536
• Institution: Nanyang Technological University

With the rapid developments in deep learning, autonomous driving become feasible and increasingly it has been applied well in real time system. As the core technology in Autonomous driving, 3D object detection gained many attentions among the AI field. Most previous researches of 3D detection are based on monocular or stereo cameras, which not perform well. Recent years, with the development of Lidar, the introduction of point cloud provides the network with accurate geometric information, leading to the remarkable progress in 3D detection files. In this paper, a comprehensive review of the 3D object detection field is delivered, including the definition, equipment, dataset, evaluation metrics and the detailed summarize of the state-of-art 3D detectors. Besides, a brief and direct taxonomy is defined to classify different methodologies. Furthermore, voxelization-based and point-based methods are analyzed particularly, comprehensive quantitative comparisons of their performances are made. Finally, a detailed discussion about the existing challenges and possible futures is illustrated.