Optical Flow for Navigation and Estimation


Computer vision presents an attractive sensor option for micro aerial vehicle (MAV) applications due to the payload and performance restrictions imposed by typical mission scenarios. Optical flow can be measured by tracking the perceived motion of feature points between successive image frames. This perceived feature-point motion yields information regarding vehicle motion through the surrounding environment.

This project involves an optimization-based approach to estimate aircraft angular rates and wind-axis angles using monocular vision. A bias in the optical-flow equations is leveraged to decouple components resulting from angular and translational motion, respectively. Attempts to resolve the ambiguity introduced by the loss of depth information are avoided through this decoupling. Additionally, estimator performance is shown to rely on proper selection of feature points used for the estimation process. Parallax measurements are used to identify features that are most likely to yield accurate state estimates. The technique is then demonstrated through simulation.

Optical flow can be used for navigation purposes as well. Perceived relative motion provides a quick indication of obstacle proximity that can be used to augment systems that employ computationally-burdensome scene reconstruction algorithms. A multi-rate planning and control strategy is investigated to study the utility and limitations of systems that integrate these disparate forms of visual information. Hence, a visual navigation system can be developed that displays a greater level of robustness to the uncertainties and computational limitations than is displyed by a system which employs either approach individually.

Optical Flow