Abstract
The introduction of the Kinect sensors by Microsoft sparked a productive period of research in the application of RGB-D (colour and depth) cameras. With the original two versions of the Kinect sensors discontinued, a number of alternatives have arisen. Many of these are based on stereo depth estimation, rather than structured light or time-of-flight ranging, and it is not clear what effect this has on techniques developed with Kinect sensors in mind. In this paper we evaluate a selection of RGB-D cameras for voxel-based RBG-D SLAM. We find that while the Kinect v2 does give less noisy reconstruction, and lower path reconstruction error, low-cost stereo sensors such as Intel’s D415 and D435 are suitable for use in this application.