A benchmark for RGB-D visual odometry, 3D reconstruction and SLAM

We introduce the Imperial College London and \nNational University of Ireland Maynooth (ICL-NUIM) dataset \nfor the evaluation of visual odometry, 3D reconstruction and \nSLAM algorithms that typically use RGB-D data. We present \na collection of handheld RGB-D camera sequences within \nsynthetically generated environments. RGB-D sequences with \nperfect ground truth poses are provided as well as a ground \ntruth surface model that enables a method of quantitatively \nevaluating the final map or surface reconstruction accuracy. \nCare has been taken to simulate typically observed real-world \nartefacts in the synthetic imagery by modelling sensor noise in \nboth RGB and depth data. While this dataset is useful for the \nevaluation of visual odometry and SLAM trajectory estimation, \nour main focus is on providing a method to benchmark the \nsurface reconstruction accuracy which to date has been missing \nin the RGB-D community despite the plethora of ground truth \nRGB-D datasets available.