Optic Flow-based Vision System for Autonomous and Collision-free Navigation of Micro Aerial Vehicles

In this work, a monocular camera-based obstacle avoidance system was designed to improve autonomous operation of quadcopters, already hindered by sensors' heavyweight and processing and energy requirements. The work underwent two stages. First, the system dynamics were modeled, linearized, and controlled using proportional and derivative (PD) controllers. Then, an optic flow-based hybrid obstacle avoidance algorithm was developed. The algorithm consisted of three approaches that account for avoiding frontal and peripheral objects, while aiming to the final position. The developed system was tested in a challenging scenario that mimicked a real forest using Webots simulator. Results demonstrated an 85% success rate of avoiding obstacles. Cases of system failure resulted due to the linearization constraint which blocked ad hoc aggressive behaviour.