Precise Landing Control of UAV Based on Binocular Visual SLAM

In this paper, we propose a precise landing algorithm for quadrotor Unmanned Aerial Vehicle(UAV) based on improved binocular visual SLAM system. The salient features of the proposed scheme are: (1) For the front-end, the advantages and disadvantages of FAST, ORB and BRISK corner extraction methods are compared, and finally the ORB algorithm with the highest accuracy and better real-time performance is adopted for the front-end corner extraction process. The descriptors of feature points are calculated by the BRIEF method, which has good rotation invariance. In order to make matching point pairs input to the back-end have higher credibility, the GMS strategy is innovatively introduced to filter the feature point pairs matched in the previous step, and the most consistent pairs are obtained. (2) In the back-end optimization part, the direct linear method in PnP is used to solve the pose between 3D and 2D images to ensure the accuracy of the output matrix. In the proposed landing algorithm, although the GPS positioning accuracy is not accurate, the deviation between the landing point and the launching point is far less than that using traditional slam framework. Finally, the accuracy and effectiveness of the method are further verified by the simulation results in the actual data set.