Autonomous robot localization with GIST method

Abstract

This thesis described an image matching module, which was designed to determine location for mobile robot. GIST, it is defined as a global feature. With a mere glance at a complex picture, an observer can seize the clutter and the variety of details. GIST can be used for this module as well. In this thesis, we brought in and validated a simple background based scene recognition algorithm for mobile robotics applications. This model can distinguish key frames from a series of input images; it can also capture the high-dimensional 'GIST' code of the scene into a low-dimensional vector.