Stata Center Frame: A Novel World Assumption for Self-Localization

Map matching is a commonly applied localization method to mobile robots. Due to the complexity of building maps and the matching task itself, many studies have adopted simplifying assumptions (geometrical and directional) of the world, such as the Manhattan World, the Atlanta World, a Mixture of Manhattan Frames and the Stata Center World. Even though the latter has flexibility to represent several environments, it has been so far limited to scene segmentation and has not yet been applied to self localization. This work explores the capabilities of the Stata Center World for self localization and further proposes a novel concept of Stata Center Frame. This assumption permits orientation estimation with one single line and position estimation by visual and positional patterns from the scene using only one monocular camera. The results show that self-localization can be achieved online with higher accuracy (average 0.12 m error) comparing to traditional techniques (0.13 m, offline).