Stable image matching for 3D reconstruction in outdoor

Stable image matching is still a pursuit goal in the communities of computer vision and graphics because of many potential applications such as 3D reconstruction, object recognition, and visual tracking. In the case of 3D reconstruction, modern structure-from-motion methods depend heavily on image-matching methods for recovering geometric models. Existing methods have tried to exploit various strategies to generate correct feature correspondences for 3D reconstruction. However, there are no efficient methods that can work well in the wild. To defend this drawback, in this paper, we design a stable image-matching method for 3D reconstruction. The method consists of feature point detection, descriptor extraction, feature matching, and matching verification. Besides, we try to choose a stable algorithm for each module of the proposed method, in which the adopted algorithms are implemented in an optimized approach for acceleration. Finally, we conduct a systematic experiment on several benchmarking datasets to assess the performance of the proposed method. Experimental results show that the proposed method has both high matching precision and fast speed.