A Fast Aerial Images Mosaic Method Based on ORB Feature and Homography Matrix

Image mosaic is of great significance for the research of computer vision. However, the environment is constantly changing over time. Accordingly, its image mosaic also needs to be regularly updated to support various applications, including comprehensive improvement of road environment, resolution enhancement and motion detection. A fundamental step for panoramic image is to register aerial images taken from different view efficiently and effectively. Aiming at the problem of low speed and low accuracy for extracting features from aerial images. An improved method based on ORB feature for image mosaic is proposed in this paper. This image mosaic process consists of four steps: Firstly, we analyze the performance and speed of the existing several popular feature extraction algorithms and propose to use ORB feature because of its more effective calculation speed. Secondly, we introduce a fast and robust feature matching strategy based on descriptors similarity by setting an appropriate threshold. Thirdly, After the matching feature set is determined, the transformation model is constructed and the model parameters are obtained, the progressive LMedS&LS algorithm is applied to eliminate false matches. Finally, the multi-band fusion algorithm is used to fuse matched images and realize panoramic mosaic. A large number of experiments demonstrated this method is efficient in the stages of feature extraction of images stitching and matching.