On user-defined region matching for augmented reality

This paper presents a study of feature-based approaches such as SIFT, SURF and ORB as a basis for markerless methodologies for creating augmented reality. In feature-based tracking methods, the target region is represented and tracked in the feature space. A discriminative and repeatable feature description is required to depict local structures according to the information from the spatial neighborhood pixel patterns. The feature representation is invariant to changes in viewing conditions such as scale, orientation and contrast. By matching the feature descriptors from different images, the same region can be detected according to the correspondence relationships between them. Using homography relations, the graphical objects can then be overlaid on the selected region which can be translated and rotated according to relationships between the matched regions.