Abstract
In this thesis, we contribute a supervised system for the recognition of roadside objects which are of interest to a navigational aid for individuals who are visually impaired. These objects of interest have known geo-location (latitude and longitude) and so the recognition of such objects along with GPS measurements would allow the navigational aid to calculate a more precise geo-location of its user than GPS alone. The proposed recognition system is a fully generalizable system and can be used to recognize other objects with only a change of training data. Additionally, we contribute a segmentation method, which is used in conjunction with the recognition system. We provide classification experiments and results using our method and show that our approach performs competitively with techniques from the literature.