Investigation of image corner features matching algorithm based on heuristic local geometric constrained strategy

摘要

The main aim of the study is to improve the performance of image matching algorithm of Scene Matching Aided Navigation System. In the paper, comer-based image matching algorithm with automatic search of homonymous corner pairs is discussed. Gaussian Low-pass Filter with different kernels according to the spatial resolution of reference image and real-time image are applied to the image in preprocessing stage to remove noise, to get over spatial resolution difference between reference image and real-time image and to enhance the repeatability of corner detection. A novel fast corner detector, which is based on SUSAN and the geometric structure analysis, is designed to extract corner features. Normalized co-correlation algorithm is applied in search of homonymous corner pairs through a small window centering corners. A heuristic local geometrically constrained strategy is employed to remove mis-matched corner pairs in initial matching stage. In the end, matched corners, in combination with a suitable polynomial algorithm, are used to match and rectify images.