Title: Collaborative Monocular SLAM with Crowd Sourced Data
Author(s): Jianzhu Huai, Grzegorz Józków, Charles Toth, Dorota A. Grejner-Brzezinska
Published in: Proceedings of the 29th International Technical Meeting of The Satellite Division of the Institute of Navigation (ION GNSS+ 2016)
September 12 - 16, 2016
Oregon Convention Center
Portland, Oregon
Pages: 1064 - 1079
Cite this article: Huai, Jianzhu, Józków, Grzegorz, Toth, Charles, Grejner-Brzezinska, Dorota A., "Collaborative Monocular SLAM with Crowd Sourced Data," Proceedings of the 29th International Technical Meeting of The Satellite Division of the Institute of Navigation (ION GNSS+ 2016), Portland, Oregon, September 2016, pp. 1064-1079.
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In
Abstract: Simultaneous localization and mapping (SLAM) is a cornerstone for a plethora of applications in robotics and augmented reality. It can be done with visual data collected by a single camera. Alternatively, if the visual data are crowd-sourced by multiple cameras, collaborative SLAM presents a more appealing solution. Its benefits are: (1) Each camera user can navigate based on the map built by other users; (2) The computation is shared by many processing units. Many researchers have looked into the collaborative mapping problem with a variety of sensors. Regarding monocular cameras, to the best of our knowledge, none of the existing studies demonstrated online operation with forward-looking cameras. To bridge this gap, this paper presents a collaborative SLAM approach with the following attributes: (1) Multiple users can localize within and work with a map merged from maps built by each user due to commonly visited areas; (2) The size of the map grows only if a new area is explored; (3) A robust loop closure approach solves for subsets of variables involved in the pose graph step by step. The proposed approach follows a client-server paradigm, the client referring to the processing unit at the user, the server being the master that can be hosted in a cloud computing framework. The communication between clients and the server in the current implementation is unidirectional, that is, the clients transmit their messages to the server, making this approach amenable to a low communication bandwidth. To validate attributes of the proposed approach, tests were conducted on the real world KITTI benchmark, and on datasets crowd-sourced by smartphones in the outdoor environment. These tests showed that even a server hosted on a consumer grade computer could process keyframes coming concurrently from several clients in real time, and create compact and accurate maps.