Monocular-based collision avoidance system for unmanned aerial vehicle

Abstract

Obstacle avoidance based on a monocular camera is a challenging task due to the lack of 3D information for Unmanned Aerial Vehicle. Recent methods based on Convolutional Neural Networks for monocular depth estimation and obstacle detection become widely used. However, collision avoidance with depth estimation usually suffers from long computational time and low avoidance success rate. A new collision avoidance system is proposed which uses monocular camera and intelligent algorithm to avoid obstacles on real time processing. Several experiments have been conducted on crowded environments with several object types. The results show outstanding performance in terms of obstacles avoidance and system response time compared to contemporary approaches. This makes the proposed approach of high potential to be integrated in crowded environments.