Design and Implementation of the UAV Reconnaissance System

Abstract

This paper designs a UAV(Unmanned Aerial Vehicle) reconnaissance system for obtaining panoramic image and relative position of small targets under simple background. The system mainly includes two parts: image stitching and object detection. The paper improves the image stitching algorithm based on the SURF. This system adds an image enhancement step to perform histogram equalization on the brightness channel of the image based on the traditional image stitching algorithm, which greatly improves the success rate of stitching. The target detection algorithm uses YOLO v3. According to the structural characteristics of the detection algorithm, the panoramic image obtained by image stitching is segmented to improve the detection accuracy. In addition, a circular cooperative target of known physical size is used in this system to roughly estimate the distance between targets in a monocular scene. The system designed in this paper has been verified by real flight experiment and achieved good results.