Application of a Nonlinear Recursive Visual-Depth Observer Using UH60 Flight Data

Abstract

This paper considers the problem of nearness (inverse of the distance to terrain) estimation from optical flow observed by a downward-looking camera attached to a helicopter undergoing known motion. A recursive nonlinear-observer framework is developed by combining online body-state estimates with pixel-based representation of optical flow for dense depth-map estimation. Lyapunov-stability analysis is considered under the assumption of bounded nearness, and global exponential convergence is demonstrated over observable regions of the image plane, which is in contrast with extended Kalman-filter-based estimation schemes that preclude guarantee of convergence. The observer estimates converge rapidly to ground-truth estimates, and the performance of the observer is shown to be similar to the extended Kalman filter, while being more computationally efficient and more robust to model error. The rapid convergence and accuracy of nearness estimation allied with the computational efficiency of the recursive observer allow for real-time implementation suitable for navigation and control.