Real-time object detection and pose estimation using stereo vision. An application for a Quadrotor MAV

This paper presents a novel strategy for object detection applied on a Quadrotor micro aerial vehicle (MAV) navigating in unknown urban environments. The Quadrotor is required to fly across a window and complete a transferring flight between an outdoor position to a final point inside a building. To achieve this goal, three main tasks must be accomplished; the first one involves the identification of the object of interest, in this case a window; the second task involves the pose estimation of the MAV w.r.t the window; and finally generating a trajectory needed to cross the window starting from a given initial point. To identify the window, a feature-based cascade classifier is implemented, which provides an extremely fast and robust method for window identification. We develop a safe path-planning method using the information provided by the GPS and the on-board inertial and stereo vision sensors. Therefore, the stereo vision system estimates the relative position w.r.t. the Quadrotor and offers egomotion estimation of the MAV for subsequent position control. Preliminary experimental results of the identification of the window and pose estimation is demonstrated through some video sequences collected from the experimental platform.