Depth Measurement-Enhanced Thermal Odometry in Visually Degraded Environments

Since thermal infrared cameras are insensitive to visual degradation caused by unknown, varying illumination, they are considered as good alternatives to the visible spectrum camera in ego-motion estimation under complex lighting conditions. However, most existing visual odometry (VO) frameworks are incompatible with thermal infrared images due to high noise, low contrast, and interruptions caused by nonuniformity correction (NUC), which exacerbates scale drift and even leads to tracking failures in motion tracking tasks. As active sensing devices, depth cameras provide depth information and mitigate scale drift, enhancing the accuracy and robustness of visual odometry. In this article, we propose a depth measurement enhanced thermal odometry by improving the direct framework for robust localization in visually degraded environments. Firstly, we introduce an image enhancement strategy to strengthen the photometric constraints between the thermal frames. Then, to precisely align depth and thermal measurements, we develop an optimization-based misalignment compensation method. Furthermore, an NUC motion recovery module is developed to address motion estimation challenges caused by NUC, utilizing sparse 2D associations to construct photometric-based 2D-3D pose estimation. Finally, the effectiveness of the proposed method is evaluated on public datasets with comparison against state-of-the-art methods.