Registration methods for RGB-D cameras accelerated on GPUs

Inhalt:
Registration of point clouds obtained from RGB-D sensors is a challenging and time demanding problem. In this paper we present the GPU implementation of different dense registration methods. Algorithms have been redesigned and optimized to maximize their performance on GPUs. A deep study has been developed in order to automatically estimate the best parameters to achieve a high speed-up and accurate results. A performance comparison with different GPUs is also provided. Accelerated registration methods can be applied to time constrain tasks like location and mapping in robotics field.