Vision-Based Recursive Estimation of Rotorcraft Obstacle Locations

Development of an onboard obstacle detection and estimation scheme for low altitude rotorcraft fight is necessary both for the development of pilot warning system and as a step toward achiving fully autonomous flight. Vision sensors provide passive sensing of obstacles, and allow a wide field of view and nearly infinite range with relatively low cat. We consider the problem of estimating the relative location of identifiable features on nearby obstacles, assuming a sequence of noisy camera images and imperfect measurements of the camera's translation and rotation. An iterated extended Kalman filter is used to provide recursive range estimation; the correspondence problem is simplified by predicting and tracking each feature's image within the Kalman filter framework. Simulation results are presented which show convergent estimates and generally successful feature point tracking. Estimation performance degrades for features near the optical axis and for accelerating motions; image tracking is also sensitive to angular rate. Our approach is also applicable to other vision-based obstacle detection and estimation problems.