Aiming at the problems of large number of redundant points, poor realtime performance and low ability to resist geometric transformations for the scene matching algorithm based on traditional local invariant feature, an unmanned aerial vehicle (UAV) scene matching algorithm based on CenSurEstar is proposed. Firstly, the CenSurEstar filter is adopted to extract the feature points in the reference image and realtime image, and then FREAK binary descriptors are generated. Secondly, the Hamming distance is taken as the similarity measurement of the feature points, and the KNearest Neighbor distance ratio method is used to extract the matching point pairs. Finally, the positioning model based on RANSAC is utilized to obtain the space geometric transformation relations, the image matching is achieved, and the latitude and longitude coordinates of the positioning points are obtained. The algorithm performance evaluation experiments show that compared with SIFT, SURF and ORB algorithms, the proposed algorithm has better robustness in dealing with various image transformations; and compared with the improved SIFT and SURF algorithms, the processing time of the proposed algorithm is greatly shortened. The positioning error of the algorithm is within 0.8 pixels, the scale error is within 0.02 times, and the rotation angle error is within 0.04 degrees. Based on the proposed algorithm, a field flight experiment was conducted. The experiment results prove that the proposed algorithm has high positioning accuracy, can adapt to the environment with less landform information, and meets the requirements of UAV vision aided navigation.