Camera + LiDAR Sensor Fusion Methods for Semantic Segmentation in Autonomous Driving : A Literature Review

Abstract

Autonomous vehicles require perception systems that are highly accurate, robust and capable of processing data in real-time to ensure reliable operation. To fulfill these requirements, autonomous systems can benefit from sensor fusion for semantic segmentation. The focus of this thesis is on the discussion about the fusion of a common sensor combination in autonomous vehicles; cameras and LiDAR. This thesis starts by giving an overview of semantic segmentation and sensor fusion in autonomous vehicles. Then follows explanations on fusion approaches (early, mid, late and asymmetric fusion) and common architecture types that serve as a basis for most of the fusion methods. The thesis ends with the literature review focused on deep learning-based fusion methods for camera and LiDAR, followed by a discussion on limitations of approaches and future directions.