A fast image matching technique for the panoramic-based localization

This paper proposes a novel image tracking technique for the image-based positioning study on the mobile device. The study uses the panorama as the indoor map to localize the user. Since the number of feature points of the panorama is much larger than feature points from the image of the mobile device, the performance of matching mobile device image with the panorama will significantly affect the efficiency of the positioning. Hence, a Sorting Hat approach is proposed to filter out uncorrelated feature pairs as early as possible. The Sorting Hat is designed according to the characteristic of BRISK which is adopted as feature detection technique in this study. However, since the coordinates of the panorama is not the same as that of mobile device image, a coordinates unification is proposed to improve the effectiveness of image matching. To further improve the image matching speed during the positioning process, a Reckon-View mechanism on the panorama is also proposed. Our experiments show that the Sorting Hat approach can decide if two feature points are correlated within 0.824ms at 99.6% confidence. Furthermore, the experiment also prove that the Reckon-View mechanism can reduce the number of feature pairs computation to two-thirds of the original.