Abstract:
The knowledge of the current location is important for navigation purposes. A person generally understands his/her location by using the knowledge of the surrounding landmarks. In the urban environment, buildings can be used to play the role of the landmarks. For this purpose the building has to be reliably and efficiently recognized by matching its images with its image model.
In this work we propose a system for fast localization using mobile phone camera imagery. It is a hierarchical approach for localization. It involves two stages. In first stage, it identifies building in the query view using wide baseline matching algorithm. Our wide baseline matching algorithm is itself a three step approach for recognition of buildings. In the first step, we use previously proposed global representation named as colour histogram. Using this representation a small number of candidate images are selected from the database. In the second step, we use shape matching technique. Using shape context candidate images from previous step are further filtered out. In the third step, images are compared by using their local feature points. We propose a novel algorithm for locating feature points of the image and a clustering model to match different regions of the image. This clustering avoids random matches and makes the matching procedure independent of number of feature points of the image. In second stage, the system identifies view point of query image. We propose novel viewpoint identification algorithm for this purpose. This system is validated by experiments carried out with database consisting of images acquired by us in different light and weather conditions by using cell phone camera and also with publicly available database ZuBuD
