GPGPU implementation of On-Line point to plane 3D data registration

The paper concerns the result of the implementation of classic point to plane 3D data registration method with an improvement based on GPGPU parallel computation. 3D data is delivered by mobile robot equipped with 3D laser measurement system for INDOOR and OUTDOOR environments. Presented empirical analysis of the implementation shows the On-Line computation capability using modern graphic processor unit NVIDIA GF 580. The implementation is a part of a project “Semantic simulation engine” composed of following modules: data registration module, semantic entities identification module, semantic simulation. The goal of the project is to deliver software tools capable to create virtual model of the environment based on robot observations and perform semantic simulation, where all virtual entities correspond to real one. Possible practical application of the project are supervision and control of robotic system and mobile robot operator training using Augmented Reality techniques. Data registration module is composed of the implementation of point to point and point to plane classic methods improved by the usage of parallel computation. In this paper it has been shown that the implementation of GPGPU point to plane registration method is accurate in INDOOR structured environment but it has some difficulties in accurate alignment in OUTDOOR environment.