Research Focus
Self-localiztion using thermal images
This study aims to develop a self-localization method using thermal images to enable robots to move accurately at night or in dark environments. The main feature of this research is the adaptation of a deep learning model trained with visible-light camera images to thermal infrared camera images. Thermal infrared cameras can capture images in the dark. Still, their images often have low contrast and unclear details, making it difficult for robots to recognize environmental features and estimate their position. To address this issue, the proposed method transfers clear feature information obtained from visible-light images into the learning process, enabling the model to extract stable environmental structures from thermal images. As a result, robots can estimate their own position in real time and perform safe navigation and mapping even in harsh outdoor environments without streetlights. In the future, this technology is expected to realize autonomous systems that operate regardless of lighting conditions and contribute to automated nighttime security and patrol tasks.