Perception and pose estimation in poorly illuminated environments is a significant challenge in mobile robot navigation. Camera-based perception algorithms require a level of dynamic range in intensity that is generally unavailable at night without external or environmental light sources. The lack of intensity information also requires pose estimation techniques to be robust to feature matching errors between neighboring images. This task will address these issues on two fronts by designing a perception sensor illuminator for a small unmanned ground vehicle (SUGV) and working to improve the performance of feature tracking algorithms in these operational environments. These advances will enhance the quality of stereo disparity and pose estimation in poorly illuminated conditions.
This work is funded by SPAWAR.
Rudranarayan Mukherjee - Jet Propulsion Laboratory