Image matching is an essential process in the mosaicking of unmanned aerial vehicle (UAV) image sets. At present, the most popular methods establish putative correspondences mainly based on local features; however, some mismatch will occur due to the repetitive pattern if only the feature similarity is employed. In this case, we propose a geometric information-assisted image matching method for UAV images. As the feature matching is restricted in pairwise geometric grid cells, some unnecessary feature similarity calculations are avoided, which is of great help to shorten the matching time. Compared to other methods, the proposed method greatly improves the matching speed without the sacrifice of accuracy, which has a practical significance for real-time remote sensing applications.

Positioning information has become one of the most important information for processing and displaying on smart mobile devices. In this paper, we propose a visual positioning method using RGB-D image on smart mobile devices. Firstly, the pose of each image in the training set is calculated through feature extraction and description, image registration, and pose map optimization. Then, in the image retrieval stage, the training set and the query set are clustered to generate the vector of local aggregated descriptors (VLAD) description vector. In order to overcome the problem that the description vector loses the image color information and improve the retrieval accuracy under different lighting conditions, the opponent color information and depth information are added to the description vector for retrieval. Finally, using the point cloud corresponding to the retrieval result image and its pose, the pose of the retrieved image is calculated by perspective-n-point (PnP) method. The results of indoor scene positioning under different illumination conditions show that the proposed method not only improves the positioning accuracy compared with the original VLAD and ORB-SLAM2, but also has high computational efficiency.

Fast and efficient mismatches filtering is significant for unmanned aerial vehicle (UAV) image matching. The traditional geometric model-based filtering is time-consuming and may deteriorate when nonrigid transform exists. To extract more reliable correspondences with less running time, filtering mismatches under the guidance of a subset of initial seed matches derived from putative matches is proposed. The method includes three steps: 1) obtain a subset of initial seed matches based on information entropy and range searching; 2) construct a seed kd-tree with initial seed matches to determine the remaining matches according to the consistency of k-nearest neighbors, and then add the true matching point to the seed kd-tree; and 3) repeat step 2 until all remaining putative matches are confirmed. Extensive experiments prove that the proposed method is competitive compared with state-of-the-art methods.