2377-3766 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LRA.2016.2516585, IEEE Robotics and Automation Letters IEEE ROBOTICS AND AUTOMATION LETTERS. PREPRINT VERSION. ACCEPTED DECEMBER, 2015 1 Accurate Continuous Sweeping Framework in Indoor Spaces with Backpack Sensor System for Applications to 3D Mapping Keonyong Lee † , Soo-Hyun Ryu † , Suyong Yeon, HyunGi Cho, ChangHyun Jun, Jaehyeon Kang, Hyunga Choi, Janghun Hyeon, Insik Baek, Woonhyung Jung, Hanul Kim and Nakju Lett Doh Abstract—In indoor environments, there exists a few distinctive indoor spaces’ features (ISFs). However, up to our knowledge, there is no algorithm that fully utilizes ISF for accurate 3D SLAM. In this paper, we suggest a sensor system that efficiently captures ISF and propose an algorithm framework that accu- rately estimates sensor’s 3D poses by utilizing ISF. Experiments conducted in six representative indoor spaces show that the accu- racy of the proposed method is better than the previous method. Furthermore, the proposed method shows robust performances in a sense that a set of adjusted parameters of the related algorithms does not need to be recalibrated as target environment changes. We also demonstrate that the proposed method not only generates 3D depth maps but also builds a dense 3D RGB-D map. Index Terms—Localization, Mapping, SLAM I. INTRODUCTION A S indoor environments are man-made artificial construc- tions, there exist a few distinctive indoor spaces’ features (ISFs) compared to natural spaces as follows. • structured: all buildings consist of ‘structure’ built by a construction company and ‘objects’ located or installed by users. • direction-segmented: according to the classification in [1], the major portion of indoor spaces can be labeled as ‘direction-segmented world’ which has distinctive direc- tional features. • plane-based: planes are the dominant geometric feature in indoor spaces. Although there have been significant improvements in SLAM, it seems like that there is no 3D algorithm that fully utilizes ISF within a well-organized framework. In [2], a general framework that uses structural properties in the feature-based graph formulation was proposed. However, this approach considers 2D SLAM and addresses its extension to 3D as a challenging future work. Except this work, other researches do not fully use of ISF but partially take advantages from some of ISF. For example, Trevor et al. [3] employed Manuscript received: August, 31, 2015; Revised November, 26, 2015; Accepted December, 17, 2015. This paper was recommended for publication by Editor Cyrill Stachniss upon evaluation of the Associate Editor and Reviewers’ comments. This work was supported by the Global Frontier R&D Program NRF-2011-0031648 and by the National Research Foundation grant of NRF-2012R1A2A2A01044957. † Those authors contributed equally. All authors are with the School of Electrical Engineering, Korea University, Republic of Korea. All correspon- dences should be addressed to nakju@korea.ac.kr. Digital Object Identifier (DOI): see top of this page. Fig. 1. The proposed backpack-type sensor system with a 3D LIDAR (Velodyne HDL-32E), an omni-directional camera (Pointgrey Ladybug2), and a 9 DOF IMU (MicroStrain 3DM-GX3-45). domain knowledge that indoor spaces consist of large walls because measurement errors of detected point features can be reduced by projecting them to the nearest wall. In [4]–[7], directional features were used to enhance the pose estimation accuracy under the Manhattan world assumption. In [8] and [9], 3D maps of indoor spaces were built by minimizing errors between two consecutive sets of planes. In this paper, we suggest a sensor system that efficiently captures ISF. Then, we propose an algorithm framework that accurately estimates sensor’s 3D poses by utilizing ISF. Regarding the sensor system, we suggest a backpack-type first- person-system composed of a 3D LIDAR (Velodyne HDL- 32E), an omni-directional camera (Pointgrey Ladybug2), and a 9 DOF IMU (MicroStrain 3DM-GX3-45) as in Fig. 1. By virtue of this LIDAR’s characteristic of the wide field-of-view (360 ◦ in horizontal direction), the long range detection (up to 80m), and the fast speed (10Hz), this system can capture ISF even when the system is under arbitrary motions. As this system continuously moves in 3D spaces, we need a 3D continuous sweeping method which estimates sensor’s trajectory by considering time differences in-between instan- taneous motions. For this purpose, we propose the algorithm framework as in Fig. 2. This framework uses the open-loop Zebedee [10] as an initial estimation module and proposes a novel way that fully utilizes ISF for a global optimization. The key idea is to employ three algorithms where each algorithm is suitable for one of ISF’s characteristics and to