Hierarchical Drone Navigation Control for Ensuring Safety with Stationary Humans

Leveraging YOLOv8's segmentation capabilities, a drone is able to detect humans in its field of view to avoid collisions using only 2D information from a monocular RGB camera. To this end, in this work, the integration of a system composed of monocular ORB-SLAM2 and YOLOv8 is presented to enable autonomous drone navigation in indoor environments with human presence. This is achieved with a hierarchical control scheme that manages two tasks simultaneously: human avoidance and position regulation. Drone navigation uses a 3D map build with the ORB-SLAM2 technique. The human detection is done using the YOLOv8 deep learning approach and it's COCO database training. The system effectiveness is verified through experimental validation, where the drone encounters a human while navigating with a fixed height towards a target position defined in the map. Control stability is ensured with the Lyapunov approach. The results obtained represent a promising safe drone navigation system in indoor environments with a stationary human.