Using Stereo Photogrammetry and Deep Learning combined with Dual Antenna RTK for UAV Position Tracking

Abstract

The bridge is a very important infrastructure structure connecting the social and economic transportation road network and the bridge inspection is a necessary method to maintain the availability and service safety of the bridge structure Through bridge monitoring the safety of the bridge can be guaranteed Using UAV to inspect bridges is one of the new methods The advantages of using UAV to detect bridges are comparative low cost high flexibility and wide adaptability However the positioning of the UAV depends on the satellite positioning signal When the UAV is under the GNSS denied environment such as under the bridge the antenna equipped on the UAV cannot receive positioning information It will cause the image taken by the UAV to lose its position in the geographic coordinate system and the UAV cannot perform automatic navigation In this study in order to monitor the geographic location of UAV during flying under the obstacle in real time a calibrated low-cost stereo-camera and a dual-frequency dual-antenna RTK system are installed on the platform During flying the image coordinates of UAV can be detected and tracked by using You Only Look Once (YOLO) Based on stereo photogrammetry technique the three-dimensional coordinates of UAV can be computed Furthermore with pre-calibrated coordinate transform parameters between stereo-camera and dual-antenna RTK system the geographic coordinates of UAV can be derived In this study we will propose a protype and evaluate its positioning accuracy under a well-controlled test-site In the accuracy assessment the difference between the geographic location of UAV and the reference value obtained by our method is worse than the expected accuracy that 30 centimeters within 20 meters observation distance The main source of error is that the baseline of the stereo camera used in the study is too short The results show that the method need to be improved before application

Date of Award	2020
Original language	English
Supervisor	Jiann-Yeou Rau (Supervisor)