A System for Vision-Based Flights in Unknown Urban Environments

Abstract

This paper presents a process to create threedimensional obstacle maps for collision avoidance, it is a part of the sense and avoid system of an unmanned helicopter. Classical robotic approaches like occupancy grid mapping and polygonal feature extraction are combined with aircraft-specific techniques and extended with new features to meet the requirements of flight scenarios. For example, the created map covers full 3-D scenarios, it is not restricted to a pre-defined area to be applicable to large outdoor environments, and the output is very compressed to be sent over network. The maps are created and updated onboard and in real-time, and it is possible to integrate a-priori knowledge and the data of other sensors or vehicles.