Large-scale localization datasets in crowded indoor spaces

Estimating the precise location of a camera using visual localization enables interesting applications such as augmented reality or navigation. This is particularly useful in indoor environments where other localization technologies, such as GNSS, fail. Indoor spaces impose interesting challenges on visual localization algorithms: occlusions due to people, textureless surfaces, viewpoint changes might cause large changes in appearance due to close structures, low light, repetitive textures, just to name a few. In order to develop robust algorithms that are able to cope with these challenges, a few indoor datasets were proposed. However, these datasets are either comparably small or do only cover a subset of the mentioned challenges. In this paper, we present 5 new indoor datasets for visual localization in challenging real-world environments. They were captured in a large shopping mall and a large metro station in Seoul, South Korea using a dedicated mapping platform consisting of 10 cameras and 2 laser scanners. In order to compute ground truth camera poses, we developed a robust LiDAR SLAM which provides initial poses that are then refined using a new structure from motion framework especially designed for this purpose. We also present a benchmark of modern visual localization algorithms on these challenging datasets. The datasets will be publicly released to foster future research.