Street searching service framework for navigation

In this paper, we propose new street searching service framework. Most location-based service uses GPS to estimate the geographical position; however, it is noisy and the GPS is not always available. To compensate this error, the visual search can be used to localize the position by comparing descriptors with geo-tagged reference DB, but it is a tedious job to collect the reference DB. To overcome this problem, we use visual descriptors around crossroads to localize the position. Combining location recognition with visual descriptor and multi-perspective panoramic street views, we provide precise and easy way of searching locations in downtown.