Volume IV-2
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., IV-2, 25-32, 2018
https://doi.org/10.5194/isprs-annals-IV-2-25-2018
© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 4.0 License.
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., IV-2, 25-32, 2018
https://doi.org/10.5194/isprs-annals-IV-2-25-2018
© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 4.0 License.

  28 May 2018

28 May 2018

LAYOUT SLAM WITH MODEL BASED LOOP CLOSURE FOR 3D INDOOR CORRIDOR RECONSTRUCTION

A. Baligh Jahromi1, G. Sohn1, J. Jung1, M. Shahbazi2, and J. Kang1 A. Baligh Jahromi et al.
  • 1GeoICT Laboratory, Department of Earth, Space Science and Engineering, York University, 4700 Keele Street, Toronto, Ontario, Canada
  • 2Department of Geomatics Engineering, University of Calgary, 2500 University Dr NW, Calgary, Alberta, Canada

Keywords: SLAM, Extended Kalman Filter, Layout Estimation, 3D Indoor Reconstruction, Model-based Matching, Loop Closure

Abstract. In this paper, we extend a recently proposed visual Simultaneous Localization and Mapping (SLAM) techniques, known as Layout SLAM, to make it robust against error accumulations, abrupt changes of camera orientation and miss-association of newly visited parts of the scene to the previously visited landmarks. To do so, we present a novel technique of loop closing based on layout model matching; i.e., both model information (topology and geometry of reconstructed models) and image information (photometric features) are used to address a loop-closure detection. The advantages of using the layout-related information in the proposed loop-closing technique are twofold. First, it imposes a metric constraint on the global map consistency and, thus, adjusts the mapping scale drifts. Second, it can reduce matching ambiguity in the context of indoor corridors, where the scene is homogenously textured and extracting sufficient amount of distinguishable point features is a challenging task. To test the impact of the proposed technique on the performance of Layout SLAM, we have performed the experiments on wide-angle videos captured by a handheld camera. This dataset was collected from the indoor corridors of a building at York University. The obtained results demonstrate that the proposed method successfully detects the instances of loops while producing very limited trajectory errors.