Developments in 3-D Object Detection for Autonomous Driving: A Review

Abstract

In recent years, 3-D object perception has emerged as a critical component in the development of autonomous driving systems, offering essential environmental awareness. As perception tasks become increasingly complex, a variety of detection techniques have been proposed, leading to diverse perspectives from both academia and industry. While numerous surveys exist, they primarily focus on specific detection methods or single-sensor approaches, lacking a broader perspective that analyzes the landscape of 3-D object perception across multiple modalities. This review provides a comprehensive and panoramic perspective by systematically summarizing and analyzing 3-D object detection methods, encompassing camera-based, light detection and ranging (LiDAR)-based, and multisensor fusion techniques. Beyond evaluating the strengths and limitations of these approaches, we examine the critical challenges encountered in real-world applications, such as synchronization issues, calibration drift, and the inherent limitations of sensor fusion. Furthermore, we explore emerging research directions, including temporal perception, 3-D occupancy grids, and cooperative perception methods that extend the perception range through collaborative communication. By offering a holistic view of the current progress and future developments in 3-D object perception, this review serves as a valuable resource for researchers and practitioners. In addition, to facilitate continuous updates on the latest advancements in the field, we have established an active repository, accessible at: https://github.com/Fishsoup0/Autonomous-Driving-Perception