Implementation, Testing, and Comparison of Feature Detection Algorithms for Micro-GPS

Abstract

Micro-GPS – proposed by Rusinkiewicz and Finkelstein – would provide the world with subcentimeter location determination, simply by finding identifying features in the textures of the ground. Micro-GPS opens up wide-ranging applications, including automated lane control for cars, robot navigation in warehouses, among many others. In this thesis, I implement and evaluate four different feature detection algorithms, to determine which – if any – are best suited to the needs of the larger Micro-GPS project. One of the most significant outcomes of this research project is the multi-step testing suite I have developed for assessing existing and future data sets. Based upon the results of the testing suite, I propose that Difference-of-Gaussians blob detection should be the primary algorithm used for the Micro-GPS project, though Canny edge detection could be used in conjunction to find appropriate parameters.