On the safe navigation problem for unmanned aircraft: Visual odometry and alignment optimizations for UAV positioning

With increasing automation of unmanned aircraft and the endeavor to fly between buildings in cities and in other occluded areas, safe navigation is essentially required but still a challenge. This paper is about the important issue of vehicle positioning in the case of satellite signal dropouts, and it presents a visual odometry method to compensate GPS positioning interruptions. The presented approach follows common triangulation principles and nonlinear optimization methods. Absolute scale is obtained by a stereo camera, although stereo is required only from time to time. In addition to the estimation of the vehicle position, the method estimates the camera alignment with respect to the vehicle, yielding a more accurate map and pose estimation. Returned vehicle poses are available in real-time with high update rates, being ready for an integration into state estimation and flight control. To demonstrate the algorithm properties, the paper incloses the evaluation of sensor data from unmanned helicopter flight tests. It shows the successful bridging of satellite positioning gaps by calculating the vehicle trajectory only by vision. Finally, the paper discusses some open issues for future work.