3D Object Detection for Autonomous Driving: A Survey

In recent years, autonomous driving has attracted significant attention. 3D object detection is a crucial component of autonomous driving systems, because it provides essential information for downstream tasks such as target tracking, path planning, and obstacle avoidance. So, it is meaningful to review existing 3D object detection methods and highlight their strengths and limitations. LiDAR and camera sensors are commonly used by autonomous vehicles to collect point clouds and images. Based on the type of sensors used, we roughly categorize 3D object detection methods into three classes: LiDAR-only method, Camera-only method and LiDAR-Camera fusion method. In this review, we further categorize each class into several subtypes, and describe the core ideas of the methods in each subtype. Furthermore, this paper critically evaluates their strengths and limitations. Finally, by using public datasets, a comparative performance analysis of the methods is conducted.