Precise Localization of a UAV with Single Vision Camera and Deep Learning

Hyeong Tae Kim, Hwangnam Kim

Abstract: This paper suggests a novel method of detecting and estimating the position of Unmanned Aerial Vehicle (UAV) with a single monocular camera only. As the leverage of UAV is keep on increasing, the related research has been extremely developed. To successfully use a UAV in a variety of missions, a precise localization technique is essential. However, there is still a lack of research to accurately measure the vehicle's present altitude. Thus, this study conducted a simple but accurate altitude measurement method using a camera. First, UAV detection is initially proceeded by using a deep learning approach. After determining that the object displayed in the image is UAV, the altitude is calculated with a distance measuring formula using the camera's Field of View (FOV). Besides, zooming, cropping, and some image processing are performed to enhance the accuracy of the altitude value. As a result, average errors of less than 5% and errors of up to 60cm were obtained, which is an improvement over previous altitude measurement techniques. This method can calibrate the altitude of the UAV immediately in a relatively inexpensive and simple way.

External IDs:dblp:conf/globecom/KimK20