TR2013-043

Manhattan Junction Catalogue for Spatial Reasoning of Indoor Scenes


    •  Ramalingam, S.; Pillai, J.K.; Jain, A.; Taguchi, Y., "Manhattan Junction Catalogue for Spatial Reasoning of Indoor Scenes", IEEE Conference on Computer Vision and Pattern Recognition (CVPR), DOI: 10.1109/CVPR.2013.394, ISSN: 1063-6919, June 2013, pp. 3065-3072.
      BibTeX Download PDF
      • @inproceedings{Ramalingam2013jun,
      • author = {Ramalingam, S. and Pillai, J.K. and Jain, A. and Taguchi, Y.},
      • title = {Manhattan Junction Catalogue for Spatial Reasoning of Indoor Scenes},
      • booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
      • year = 2013,
      • pages = {3065--3072},
      • month = jun,
      • doi = {10.1109/CVPR.2013.394},
      • issn = {1063-6919},
      • url = {http://www.merl.com/publications/TR2013-043}
      • }
  • MERL Contact:
  • Research Area:

    Computer Vision


TR Image
Figure 1. A living room with several junctions of types L, T, Y,X and W. We present a novel method to detect these junctions and demonstrate that they are discriminative features in recovering the spatial layout of an indoor scene.

Junctions are strong cues for understanding the geometry of a scene. In this paper, we consider the problem of detecting junctions and using them for recovering the spatial layout of an indoor scene. Junction detection has always been challenging due to missing and spurious lines. We work in a constrained Manhattan world setting where the junctions are formed by only line segments along the three principal orthogonal directions. Junctions can be classified into several categories based on the number and orientations of the incident line segments. We provide a simple and efficient voting scheme to detect and classify these junctions in real images. Indoor scenes are typically modeled as cuboids and we formulate the problem of the layout estimation as an inference problem in a conditional random field. Our formulation allows the incorporation of junction features and the training is done using structured prediction. We outperform other single view geometry estimation methods on standard datasets.