Master Thesis: Motion Estimation based on Thermal Inertial Odometry

Abstract

Autonomous odometry of aerial robots in environments that are visually degraded, such as dark evening or smoke-filled, is always hard and challenging. However, a thermal camera could be one potential solution since it performs in the long-wave infrared spectrum and is thus not affected by the scene illumination changes or certain obscurants. Inspired by this fact, this thesis proposes a thermal-inertial odometry (TIO) framework based on the visual-inertial odometry (VIO) framework considering the compatibility between those two. The TIO is tailored to utilize the full-size (16-Bit) thermal image data in feature detection and feature tracking. In parallel, a rescaled image (8-Bit) is used for loop-closure detection. The proposed TIO framework was validated with extensive datasets, including daytime and night-time scenes.