Evaluating Egomotion and Structure-from-Motion Approaches Using the TUM RGB-D Benchmark
J. Sturm, Wolfram Burgard, Daniel Cremers · 2012
Abstract — In this paper, we present the TUM RGB-D benchmark for visual odometry and SLAM evaluation and report on the first use-cases and users of it outside our own group. The benchmark contains a large set of image sequences recorded from a Microsoft Kinect associated with highly accurate and time-synchronized ground truth camera poses from an external motion capture system. The dataset consists in total of 39 sequences that were recorded in different environments and cover a large variety of scenes and camera motions. In this work, we discuss and briefly summarize the evaluation results of the first users from outside our group. Our goal with this analysis is to better understand (1) how other researcher use our dataset to date and (2) how to improve it further in the future. I.