RGBD sensors correction with Gaussian process regression

Abstract

In this work, we present a novel method to accurately calibrate active depth sensors such as the Microsoft Kinect. Our approach is based on the Gaussian Process Regression (GPR). It is applied after the standard calibration and it is particularly useful for the aging depth cameras. The latter, were proven to suffer from inaccuracies that cannot be fixed with the standard pinhole camera calibration procedure. Experimental results show the weaknesses of this naïve calibration and the corrective of effect our algorithm. We further justify the possibility to extend the same approach to any other type of cameras with similar characteristics.