A Heterogeneous Sensor Fusion Framework for Obstacle Detection in Piloted UAVs

Abstract

Teleoperation of Unmanned Aerial Vehicle (UAV) is a demanding task that requires skill and experience. For the most part, commercial-grade UAVs are still manually piloted. Some form of obstacle detection capability is desired in UAVs to minimize the chance of collisions and to ensure safety to human lives and properties. This thesis presents a heterogeneous sensor fusion framework for obstacle detection using complementary sensors, a monocular visual camera, and distance sensors to detect obstacles. The approach focuses on obstacles at low altitude, such as static obstacles with a large surface area and thin obstacles such as cables. The fusion of inputs is performed using fuzzy logic. The warning alerts to the pilots are sent using graphical and auditory signal methods when an obstacle is encountered. The evaluation was conducted using the simulation platform Microsoft AirSim. The approach detects thin obstacles, static large obstacles, and thin obstacles with a static obstacle in the background successfully. A case study was also conducted involving a human subject to obtain qualitative evaluation. Results obtained shows that the proposed approach has a great potential in the UAV obstacle detection. The proposed framework and the evaluation results are the contributions of this work. The thesis discusses the framework's limitations and provides an overview of aspects that should be focused on when the approach is extended and implemented for a real hardware platform.