Perpustakaan Digital ITB

In this thesis, a method to create an Autonomous Landing System for use in Naval-based UAV is formulated by incorporating a deep-learning computer vision model based on YOLOv4 combined with a distance estimator to deduce a landing coordinate on a UAV equipped with a camera and an embedded computer. Images from the camera is processed using the developed vision model, outputting a bounding box that are then calculated using stadiametric rangefinding to measure distance. The attitudes of the camera as well as the UAV is combined with the measured distance, outputting a coordinate that can then be fed to the UAV????s Flight Controller. During testing, the system starts measuring distance to the landing ship from 1.4km away, with an average of 10% error in distance in ideal conditions. Performance in inclement weather is also tested, only being able to start measuring from around 400m with similar error rates afterwards