High speed 3-D registration using GPU

Abstract

This paper describes high speed 3D object recognition based on DAI (depth aspect image) matching and M-ICP (modified iterative closest point). We regards GPU(graphic processing units) as coprocessor which are capable of computation for general purpose. We proposed 3D object recognition method which consists of two step pose estimation and positioning, i.e. the DAI matching for coarse step and HM-ICP (hierarchical M-ICP) for fine one Our method on GPU which has remarkable performance for parallel computation. The experimental results show the effectiveness of our method. This method can process 2 or 3 times faster than the original one, although the calculation amount of this method is at least 20 times bigger than the original one. Additionally, its processing time is more stabler than original method.