Robot Localization by Data Integration of Multiple Thermal Cameras in Low-Light Environment

Abstract

A method is proposed for interpolating pose information by integrating data from multiple thermal cameras when a global navigation satellite system temporarily experiences a decrease in accuracy. When temperature information obtained from thermal cameras is visualized, a two-stage temperature range restriction is applied to focus only on areas with temperature variations, making conversion into clearer images possible. To compensate for the narrow field of view of thermal cameras, multiple thermal cameras are oriented in different directions. Pose estimation is performed with each camera, and the estimation results of one camera are interpolated using those of other cameras based on reliability derived from predicted values of the camera pose. Experimental results obtained in a low-light nighttime environment demonstrate that the proposed method achieves higher pose estimation accuracy than other state-of-the-art methods.