Thermal-to-Depth Driven Safe Navigation for UAVs in Degraded Environments

Hürkan Şahin, Van Huyen Dang, Erdal Kayacan

Keywords: Thermal imaging, Thermal-to-depth estimation, Recurrent neural networks, UAV, SLAM, Reinforcement learning, Reward engineering

Abstract:

We propose an end-to-end thermal navigation
framework for UAVs operating in GPS-denied and visually
degraded environments. By coupling monocular thermal-todepth
estimation with depth-based safe navigation, the framework
allows predicted depth from non-radiometric thermal
images to be used directly for policy learning and control.
In addition, a depth-driven reward formulation is designed
to encourage safer navigation behavior. Results on a custom
dataset demonstrate substantially improved depth estimation
accuracy, supporting the promise of the proposed approach for
thermal UAV navigation in challenging environments.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 9