Vision-Based SLAM System for Small UAVs in GPS-Denied Environments

Vision-based unmanned aerial vehicle (UAV) control is a core technology in global positioning system (GPS)-denied environments. However, small UAVs have difficulty processing image data in real time because of their limited payload capacity and relatively weak processor. Development of an image-processing system that is suitable for small UAVs is therefore needed. In this paper, an approach for a vision-based simultaneous localization and mapping (SLAM) system for small UAVs is suggested. For real-time localization, a method wherein Kanade-Lucas-Tomasi-based localization and tracking are performed within small UAVs in real time is adopted. At the same time, scale-invariant feature transform–based mapping and more accurate localization are performed at the ground-control station. A vision-based three-dimensional map-building method for small UAVs using a monocular camera is also proposed. The proposed method first extracts straight-line information from images and calculates their equations in three-dimensional space. It then constructs plane information from line information.