Geo-Localization Based Scene Recognition

GPS information has been applied to many fields, especially the navigation systems. With the popularity of smart mobiles, we proposed a novel method for navigation which combines mobile GPS location information with scene recognition. In our approach, three steps of filters are used. Firstly, we utilize GPS information to discard a large number of images that are not nearby the query image. Secondly, local sensitive hash (LSH) is employed to achieve dimension reduction and hamming distance is adopted to filter images by the threshold value. For the drawback of BOW discarding spatial information, we adopt RANSAC algorithm to do the spatial verification and finish the third filter. Finally, voting-based method is employed to calculate the similarity of images. Our system returns the corresponding information of the query image. The experiment results demonstrate substantial improvements in query efficiency and precision.