The invention discloses a super-large scale image feature points matching method and system thereof. The method comprises the following steps: firstly, image matching pairs are obtained by caring out image near neighbor search; images are acted as nodes, the edges of constitution among the image neighbors form an undigraph, the undigraph is caring out priority sorting of breadth, then the images and image pairs are obtained; the feature information of the images are sorted again according to the sorting result, and the feature information of the images is divided and stored as binary files; the binary files with the feature information are read according to the order of sequence; according to the images after being sorted, the characteristics match is being carried out in sequence, and the subsequent useless feature information is released in time; the feature information is iteratively read and the feature matching is being carried out until all images finish matching. The method and system has the advantages that one read and local full both of image feature points in the process of image matching are ensured, memory is not spilled and at the same moment, the efficiency of algorithm is ensured.