Floor segmentation algorithm for indoor vision/inertial integrated navigation

Abstract

Navigation is a key technology for autonomous robots,which makes them movable in an unknown environment.To tackle the difficulty of building indoor navigation map for inertial navigation systems,a new map building method for inertial/visual navigation is proposed.By limiting robot's movement within the floor areas,the global navigation map is generated from a bird-view image.An algorithm of automatic floor segmentation is proposed,which employs principal component analysis to implement dimension reduction for local color features and adopts clustering analysis to realize floor segmentation automatically.Finally,an indoor bird-view image database is built to evaluate the algorithm.The algorithm gets the worst performance,75% averaged accurate segmentation rate,on the fourth group images because illumination reflection is found in the images.Average accurate segmentation rates on other groups are around 85%.Thus,the preprocessing algorithms,such as illumination refection detection,can help to improve the performance of the algorithm.