Partitioned environments for visual localization

Abstract

In this paper, we present a novel sequential visual localization algorithm in partitioned route. The algorithm utilizes Monte-Carlo for accurate visual localization. Partitioning the route into several regions produces higher accuracy along with lower computational cost. Each of the regions is represented using independent maps. We use bag-of-words for visual mapping of the environment. In addition, it shows smooth transition when the robot moves from one region to another. Experiments are carried out using data collected from crowded roads. Results show that this method is superior to previous attempts that looked into localization in crowded outdoor environments.