Real time vision-based autonomous precision landing system for UAV airborne processor

Autonomous precision landing of unmanned aerial vehicle can be widely used in ocean inspection, mine inspection and other industries. Vision based autonomous precision landing research can be used in global positioning system (GPS) denied area with low cost and high accuracy which has become a hot research spot. Previous vision-based methods cannot cope with complex environmental changes while landing and can not achieve real time in airborne processor. A lightweight detection model named Myolo is designed to detect landmark. Fast contour optimization algorithm is proposed to further enhance location results. Experimental results show that Myolo can achieve 16.1 FPS which is 23 times faster than faster-RCNN and 5 times faster than yolo. After using fast local contour optimization algorithm, Myolo's position accuracy is increased to 4.2 pixel, 12.5% higher than yolo and 50.6% higher than tiny-yolo. Myolo can significantly outperforms state-of-the-art object detection model in terms of both accuracy and speed in complex landing environment.