Image matching techniques for vision-based indoor navigation systems: a 3D map-based approach1

摘要

In this paper, we give an overview of image matching techniques for variousrnvision-based navigation systems: stereo vision, structure from motion and mapbasedrnapproach. Focused on map-based approach, which generally uses featurebasedrnmatching for mapping and localisation, and based on our early developedrnsystem, a performance analysis has been carried out and three major problemsrnhave been identified: inadequate geo-referencing and imaging geometry forrnmapping, vulnerability to drastic viewpoint changes and big percentage ofrnmismatches at positioning. We introduce multi-image matching for mapping. Byrnusing affine-scale-invariant feature transform for viewpoint changes, the majorrnimprovement takes place on the epoch with large viewpoint changes. In order torndeal with mismatches that were unable to be removed by Random SamplernConsensus (RANSAC), we propose new method that use cross-correlationrninformation to evaluate the quality of homography model and select the properrnone. The conducted experiments have proved that such an approach can reducernthe chances of mismatches being included by RANSAC and final positioningrnaccuracy can be improved.