Semantic Image Segmentation-Based Safe Landing Point Determination Method for VTOL UAVs
Sanghyeon Kim · Journal of Institute of Control Robotics and Systems · 2022
This paper presents a safe landing point determination method for vertical take-off and landing (VTOL) unmanned aerial vehicles (UAVs) with a downward-looking camera. The semantic image segmentation method was applied to determine whether the VTOL UAV could land safely at the designated landing site during the final stage of landing. A deep neural network with a U-Net structure was constructed for the semantic image segmentation, and a deep neural network was trained to recognize the landing site and any obstacles. The dataset used for training the deep neural network was collected through efficient labeling rules. The deep neural network of the U-Net structure outputs the area of the landing field and the area of the obstacle in units of pixels. Based on this output data, the area occupied by the obstacle in the area of the landing site is calculated to determine whether to land. Finally, the performance of the safe landing point determination method was verified using a landing pad image taken from the downward-looking camera.