UAV Motion Estimation Using Hybrid Stereoscopic Vision

Motion and velocity are essential parameters for an Unmanned Aerial Vehicle (UAV) during critical maneuvers such as landing or take-off. In this paper, we present a hybrid stereoscopic rig made of a fisheye and a perspective cameras for motion estimation. The rotation and translation are estimated by a decoupling. The fisheye view contributes to determine the orientation and the attitude while the perspective view contributes to approximate the scale of the translation. Then, the calibrated stereo rig is used to estimate the altitude. While classical methods are generally based on feature matching between cameras, we propose in this paper an algorithm which tracks and exploits points in each view independently and filters the motion by Kalman filtering. Tracked points in each view are classed in two types: points located on the ground plane, which altitude is known and environment points which altitude is not known. Then, motion can be estimated robustly using the 2 points algorithm followed by a Kalman filter. We show that this approach is robust and accurate, and presents low sensitivity to noise by using the hybrid rig and Kalman filter.