Automatic Adjustment of Photometric Camera Parameters to Improve Visual Motion Estimation

Cameras are a fundamental component of modern robotic systems. As robots have become relied upon for safety-critical tasks, the need for robust sensing is apparent. Cameras have a major limitation, compared to other sensors such as LIDAR, in high-dynamic-range environments where lighting conditions rapidly change. These changes can cause visual navigation algorithms to struggle and, in some cases, fail in instances where images become severely under- or overexposed. The aim of this research is to improve the performance of visual motion estimation algorithms in challenging environments through the adjustment of onboard camera parameters. An estimator will be developed that will adjust onboard camera parameters in an on-the-fly fashion to optimize for an image metric tied directly into the performance of the VO pipeline. The performance of this estimator will be compared to standard auto-tuning algorithms employed by off-the-shelf machine vision cameras as well as previous approaches that employ heuristic image metrics.

Faculty Supervisor:

Jonathan Kelly

Student:

Justin Tomasi

Partner:

Thales Canada Inc.

Discipline:

Aerospace studies

Sector:

University:

University of Toronto

Program: