3D perception and object detection in different weather conditions for autonomous driving

Abstract

3D object detection plays a significant role in Autonomous Driving (AD) tasks, enabling vehicles to accurately perceive surroundings with higher accuracy due to the availability of depth information, which is a limitation in traditional 2D object detection. It provides detailed spatial information like position, size, and orientation of objects, which is critical for the downstream function of AD, like motion prediction, collision avoidance, and path planning. This capability is crucial for a dynamic and complex environment where objects are partially occluded, moving or cluttered; 3D object detection provides more situational awareness in comparison to 2D object detection.

The key research question is how 3D object detection at an intersection can be enhanced for AD applications, which potentially improves overall 3D perception in different weather conditions. 3D object detection is a key element of 3D perception for AD systems. Any improvement in 3D object detection can directly lead to performance enhancement in all downstream pipelines. Traffic at the intersection produces complex scenarios. An egocentric perception sensor suite of an individual vehicle cannot capture the complete view of the intersection. A complete and accurate view from infrastructure-based sensors that can perform in different weather conditions can simplify the problem.