Multi-feature fusion stereo matching method of adaptive region

The invention provides a semi-global binocular stereo matching method, and adopts a stereo matching algorithm of a minimum spanning tree based on multi-feature information fusion. According to the method, the mode that a single cost function is adopted for the whole image in cost calculation in a traditional algorithm is abandoned, the strong texture area, the weak texture area and the texture-free area of the image are distinguished before cost calculation, then appropriate cost functions are constructed for different areas for calculation, and the adaptability of the method in different environments is enhanced. The improved minimum spanning tree algorithm can construct the undirected graph into a more accurate tree structure, so that the relation between pixel points in the graph is clearer, and the mismatching rate of binocular vision stereo matching is reduced. The improved Rank transform is introduced, the noise immunity of the original Rank transform is improved, and the problem that the matching preci