SVD-SURF Based Fast And Robust Scene Matching Algorithm

For position parameter estimation of aircraft, this dissertation presents a SVD-SURF-based wide-baseline robust scene matching algorithm. Firstly, we build SURF scale space based on the singular value features of real-time image and reference image. By using fast Hessian matrix maximum values, 64-dimensional SURF feature descriptors of aerial images wre calculated. Then, based on the trace of the Hessian matrix, we complete feature points matching. Finally, by using RANSAC algorithm, we removed the outliers for accurate position estimation. Experiments with two real aerial image sequences show that this algorithm is robust to image rotation, scale transformation and noise.