Study of vision based pose estimation for UAV

Abstract

For UAV navigation Global Positioning System (GPS) is one of the most used sensors. It is very reliable but is vulnerable to interference and also cannot be used in scenarios where GPS is not available e.g. indoors. Due to developments in the field of digital image processing camera can be used effectively as a standalone or collaborative sensor for UAV navigation. In our approach we have developed an algorithm that uses monocular vision to estimate the position of UAV by using feature information between two consecutive frames. No prior information is needed to make an estimate. The algorithm estimates the position of UAV by matching features between two frames and then using this information to construct a homography matrix. The homography matrix is then decomposed to get translation of the UAV. A simulation was constructed and the proposed algorithm was studied. Upon addition of noise the error of algorithm increased, in order to minimize this error the matched feature points were refined using RANSAC and LMedS algorithm. A comparison study was done in order to check which of the two can be implemented to effectively filter out noisy features or mismatches. Finally indoor and outdoor experiments were conducted to see performance of proposed algorithm.