Autonomous Vehicle Perception Using Monocular Thermal Imaging Cameras

Abstract

This paper investigates the integration of monocular thermal imaging cameras into autonomous vehicles to address challenges faced by conventional sensors in adverse conditions. This study explores the performance of thermal cameras in low-light, darkness, and adverse weather, aiming to enhance safety and efficiency. To achieve that, the research proposes the development of machine learning algorithms tailored for interpreting thermal image. These algorithms would facilitate accurate object detection and road segmentation, overcoming limitations of conventional algorithms. The goal is to create a system that generates a bird's-eye view spatial map, highlighting drivable regions and detected objects. Comparative analysis shows that the proposed thermal camera architecture competes favorably, particularly in adverse conditions, compared to commonly used imaging sensors and architectures.