Flight Test Validation of Collision Avoidance System for a Multicopter using Stereoscopic Vision

This paper presents the flight test validation of collision avoidance system for a multicopter using stereovision technique. Stereoscopic vision offers an affordable, accessible, and effective method of collision detection. Using a ZED stereoscopic camera, depth maps were generated using image disparity to calculate the distance between the camera and obstacles in its view. An NVIDIA Jetson TX1 computer aboard the multicopter was used to process the depth maps to detect any obstacles in its flight path. The algorithm partitions the continuously updated depth maps in equal sections to determine the section containing obstacles. Once this section is chosen, a new vector is calculated to determine the “path of least resistance” and communicated to the onboard Pixhawk flight controller using DroneKit. Using these commands, the UAS maneuvers to safely correct its course and avoid the obstacle autonomously. Flight test objectives, procedures, and results are presented.