Markerless Outdoor Localisation Based on SIFT Descriptors for Mobile Applications
Résumé
This study proposes augmented reality from mobile devices based on SIFT (Scale Invariant Feature Transform) features for markerless outdoor augmented reality application. The proposed application is navigation help in a city. These SIFT features are projected on a digital model of the building façades of the square to obtain 3D co-ordinates for each feature point. The algorithms implemented calculate the camera pose for frame of a video from 3D-2D point correspondences between features extracted in the current video frame and points in the reference dataset. The algorithms were successfully tested on video films of city squares. Although they do not operate in real-time, they are capable of a correct pose estimation and projection of artificial data into the scene. In case of a loss of track, the algorithms recover automatically. The study shows the potential of SIFT features for purely image based markerless outdoor augmented reality applications. This study takes place in the MoSAIC project.