Vision-based safe autonomous UAV landing with panoramic sensors

Abstract

The remarkable growth of unmanned aerial vehicles (UAVs) has also raised concerns about safety measures during their missions. To advance towards safer autonomous aerial robots, this thesis strives to develop a safe autonomous UAV landing solution, a vital part of every UAV operation. The project proposes a vision-based framework for monitoring the landing area by leveraging the omnidirectional view of a single panoramic camera pointing upwards to detect and localize any person within the landing zone. Then, it sends this information to approaching UAVs to either hover and wait or adaptively search for a more optimal position to land themselves. We utilize and fine-tune the YOLOv7 object detection model, an XGBooxt model for localizing nearby people, and the open-source ROS and PX4 frameworks for communications and drone control. We present both simulation and real-world indoor experimental results to demonstrate the capability of our methods.