Visibility-based modelling and control for network-based robotics

摘要

We present an algorithm to model 3D workspace and to understand test scene for navigation or human computer interaction in network-based mobile robot. This was done by line-based modelling and recognition algorithm. Line-based recognition using 3D lines has been tried by many researchers however its reliability still needs improvement due to ambiguity of 3D line feature information from original images. To improve the outcome, we approach firstly to find real planes using the given 3D lines and then to implement recognition process. The methods we use are principle component analysis (PCA), plane sweep, visibility test, and iterative closest point (ICP). During the implementation, we also use 3D map information for localization. We apply this algorithm to real test scene images and to find out our result can be useful to identify doors or walls in indoor environment with better efficiency.