Stereo visual system for autonomous air vehicle navigation

We present a system to estimate the altitude and motion of an aerial vehicle using a stereo visual system. The system has been initially tested on a ground robot and the novelty lays on its application and robustness validation in an UAV, where vibrations and rapid environmental changes take place. The two main functionalities are height estimation and visual odometry. The system first detects and tracks salient points in the scene. Depth to the plane containing the features is calculated matching features between left and right images then using the disparity principle. Motion is recovered tracking pixels from one frame to the next one finding its visual displacement and resolving camera rotation and translation by a least-square method. We present results from different experimental trials on the two platforms comparing and discussing the results regarding the trajectories calculated by the visual odometry and the onboard helicopter state estimation.