Vision-Based Global Positioning System Using Improved GMS Algorithm for a UAV

Fast and accurate localization for unmanned aerial vehicle (UAV) navigation could avoid hazards when GPS is unavailable. A vision-based global positioning system is developed for UAV by means of satellite images from Google Map as reference. In this system, an ultra-robust and fast feature correspondence algorithm, called Grid-based motion statistics (GMS), is utilized for scene matching. GMS’s performance is comparable to the techniques many orders of magnitude slower, and it maintains high speed as the fast algorithms. A Least Median Square improved GMS algorithm (LMedS-GMS) is developed which is employed to solve the matching failure of the original GMS in challenging scenarios. Moreover, a robust filtering localization approach combining the random sample consensus algorithm (RANSAC) and the proposed LMedS-GMS is designed for locating the position of an UAV. Finally, a vision-based global positioning architecture is proposed using this method with the altitude and direction information from the airborne sensors. Experimental results based on offline data demonstrate that the proposed algorithm is superior to state-of-the-art methods in the accuracy and real-time performance.