Student Research

Collision Avoidance System for Multicopter UAVs using Optical Flow

The world has seen a rapid development in the unmanned aerial vehicles (UAV) technology due to the advantages they offer over conventional manned aircraft. They are simpler and cheaper than manned aircraft, and pose no threat to human operators. Increased level of autonomy can help pave the way for their mass integration into the National Airspace System for widespread applications. The focus of this presentation will be on collision avoidance system for multicopter UAVs for autonomous navigation around obstacles using optical flow as the primary machine vision. Optical flow is a method that is used to detect the motion of pixels between pairs of images. If clusters of like-colored pixels of an object move in a similar direction, the object is in relative motion. Optical flow, if sufficiently validated in flight tests, can help detect obstacles in the UAVs' flight paths using a simple camera such as a webcam, and is significantly cheaper than other methods of obstacle detection. This research uses Gunnar-Farneback algorithm to estimate the direction and speed of moving objects. An S1000 Octocopter is used as a test platform for the project. The vehicle is equipped with a Pixhawk 2 flight controller and Ardupilot software. A Jetson TX2 onboard computer is used to process the optical flow and collision avoidance algorithms. The developed algorithm is tested in multiple flight tests in a flight test site with natural obstacles present. Prior to flight tests, the algorithm was tested in simulation. Simulation and flight test results will be shown.