On scene matching for UAV based on spectral graph and relaxation iteration method

Scene matching is one of the key issues of visual aided navigation in UAVs, whose performance is closely related to image registration. According to the requirements of image registration in visual aided navigation, this article provides a novel algorithm with graph spectral method. Through the study of spectrum diagram in the form of the algebra, the distribution in the feature space is analytical in image structure. Firstly, a comparability matrix based on the graph spectral theory is created. Secondly, the characteristic similarity measure is introduced. And then, the comprehensive similarity matrix for singular value decomposition and bleaching process is created. Finally, the refreshing accurate matching probabilities are gained by the relaxation method to judge whether the image is matched. Experiments show that the novel algorithm offers better performance under the disturbance of rotation and scale.