Abstract: This paper proposes a virtual navigation system based on panoramic images. In order to stitch the images automatically from different perspectives, first extracts feature points from the scale space of the images, and assigns the main orientation. Feature vector can be calculated based on neighborhood information, and initial feature points matching can be calculated by using the distance ratio of the nearest neighbor feature point and the next nearest neighbor feature point. Feature matching refinement can be accomplished by using the RANSAC (Random Sample Consensus) algorithm, while calculating transformation parameters between the two images. After completion of image stitching, different perspective views of the scene can be obtained by reprojection algorithm, and gives user an immersive experience.

Key words: virtual navigation, image stitching, feature matching, robust estimation