Abstract:

This report describes a preliminary study on the problem of matching a live camera image of an urban scene to a small set of images, retrieved for the camera\'s current geographical location from a database, to find the most similar view. The approach is feature-based, first detecting point features in the images, then identifying potential point matches based on feature similarity, and finally applying the fundamental matrix constraint to retain good matches and remove poor matches. Results of image matching are shown for scenes under the same and varying illumination.