Stereo vision based obstacle collision avoidance for a quadrotor using ellipsoidal bounding box and hierarchical clustering

Abstract

A collision avoidance algorithm for multiple unknown static obstacles is proposed for a quadrotor system. A stereo vision system with a limited field of view and sensing range is assumed to be mounted on the quadrotor to obtain obstacle information. An ellipsoid is chosen as a circumscribed bounding box containing the obtained obstacle data points, which can be determined by solving a convex optimization problem. An affine transformation is used to form a collision cone consisting of straight lines tangent to the ellipsoid. A collision condition is examined using the collision cone and velocity vector of the quadrotor. The hierarchical clustering method is proposed to address multiple obstacles, and the bounding boxes are updated using the clusters. Numerical simulations are performed to demonstrate the performance of the proposed collision avoidance algorithm.