High-speed vision-based autonomous indoor navigation of a quadcopter

A monocular vision-based approach to indoor autonomous navigation for an off-the-shelf, low-cost Micro Air Vehicle (MAV) quadcopter is presented. Our approach is fully automated and relies on the extraction and analysis of the visual contours of the surrounding physical environment to successfully steer the MAV in hallways and to turn at intersections. All image analysis and processing necessary for deriving and controlling the flight trajectory take place off-board on a system external to the MAV. This permits the use of sophisticated multithreaded real-time algorithms that do not limit the speed of the drone. Furthermore, due to the elimination of on-board processing and possibly the use of additional sensors for realizing such autonomy, stock drones can be used and flight times realized on a single charge remain unaffected. We describe a prototype implementation on a Parrot AR.Drone quadcopter and initial results for autonomous navigation in a structured indoor environment that establishes the viability of the approach.