I. Introduction
Ego-motion estimation and mapping are crucial for an autonomous vehicle. Much of robotics research has focused on imaging sensors and LiDAR [1]–[5] for navigating environments without a global positioning system (GPS). Conventional RGB cameras operating under the human visible spectrum hinder in a challenging environment such as fog, dust, and complete darkness. Recently, thermal-infrared cameras have been highlighted by studies for their perceptual capability beyond the visible spectrum and their robustness regarding environmental changes.