Geolocation on the iPhone by automatic street sign reading

In any country, roads are marked with signs. In particular, street signs are a type of road sign used to display the names of streets. Mobile devices such as smart phones are now powerful enough to serve as platforms for computer vision applications. These mobile devices are commonly equipped with high resolution cameras and increasingly with internal GPS for geolocation.We propose an accurate method of geolocation which detects, segments and reads street signs in complex natural scenes from an iPhone image. Street signs are detected using image segmentation based on edge detection and contour detection techniques. Sign candidates are selected using several heuristics, and then partitioned into individual characters. The letters are recognised using a nearest-neighbour algorithm for Optical Character Recognition (OCR). The street names finally are passed to a GeoCoding API to display a street map of the users location. We evaluate the system for a real word data set and achive 75% sign detection accuracy, 91% OCR accuracy and 55% total geolocation accuracy.