SCIENCE CHINA Information Sciences, Volume 63 , Issue 1 : 119102(2020) https://doi.org/10.1007/s11432-019-9867-9

Effective two-view line segment reconstruction based on structure priors

More info
  • ReceivedJan 25, 2019
  • AcceptedApr 1, 2019
  • PublishedSep 12, 2019


There is no abstract available for this article.


This work was supported in part by National Key RD Program of China (Grant No. 2016YFB0502002), National Natural Science Foundation of China (Grant Nos. 61872361, 61772444, 61421004, 61873264), National Science Foundation for Young Scientists of China (Grant No. 61703397), Natural Science Foundation of Henan Province (Grant No. 162300410347), and Key Scientific and Technological Project of Henan Province (Grant Nos. 162102310589, 192102210279).


[1] von Gioi R G, Jakubowicz J, Morel J M. LSD: a fast line segment detector with a false detection control.. IEEE Trans Pattern Anal Mach Intell, 2010, 32: 722-732 CrossRef PubMed Google Scholar

[2] Li K, Yao J. Line segment matching and reconstruction via exploiting coplanar cues. ISPRS J Photogrammetry Remote Sens, 2017, 125: 33-49 CrossRef ADS Google Scholar

[3] Zhao H S, Shi J P, Qi X J, et al. Pyramid scene parsing network. In: Proceedings of Computer Vision and Pattern Recognition, Honolulu, 2017. 6230--6239. Google Scholar

[4] Tola E, Lepetit V, Fua P. DAISY: an efficient dense descriptor applied to wide-baseline stereo.. IEEE Trans Pattern Anal Mach Intell, 2010, 32: 815-830 CrossRef PubMed Google Scholar

[5] Huang X F. Cooperative optimization for energy minimization: a case study of stereo matching. 2007,. arXiv Google Scholar

  • Figure 1

    (Color online) 3D LS reconstruction. (a) Initial detected LSs (top) and fewer matched LSs (bottom) by the method of [2]; (b) the neighboring system constructed by the Delaunay triangulation method (the yellow square and red circles denote the current LS and its neighboring LSs, respectively); (c) initial non-matching LS $\#$305 (yellow square dot) and its neighboring LSs (red circle dots) in $I_1$; (d) by finding co-planes, LS $\#$(305) in $I_2$ is correctly matched with LS $\#$305 in $I_1$, and initial falsely matched LSs (thin LSs) in $I_2$ (with $\#$311 and $\#$314 in $I_1$) are corrected as $\#$(311) and $\#$(314) (thick LSs); (e) 3D LSs reconstructed from initial LS matches; (f) the top-view of (e); (g) 3D LSs produced by our method (different colors denote the 3D LSs in different planes); (h) the top-view of (g).


    Algorithm 1 3D LS inference using co-plane cues

    Input: initial LS matches ${\cal~L}$.

    Output: inferred and corrected LS matches ${{\cal~L}^{\rm{*}}}$.

    1: Generate seed planes $S$ from $l~\in~{\cal~M}$ and its neighboring

    LS set $N(l)$; add reliable LS matches to ${{\cal~L}^{\rm{*}}}$.

    2: Explore dominant scene planes ${\cal~H}$.

    3: For each seed plane $P(~l~)~\in~S$,

    3.1: Generate its plane family ${{\cal~H}^{\rm{*}}}:\{~{{h_i}}~\}$ from the

    associated plane $h~\in~{\cal~H}$;

    3.2: Assign the optimal plane ${h_i}$ to unvisited LSs in


    3.3: Add reliable LS matches to ${{\cal~L}^{\rm{*}}}$ and update ${\cal~H}$;

    3.4: Add the LS set to $S$ if any of its component LSs

    is assigned to the optimal plane;

    3.5: Update the neighbors of LS $l$.

    4: Output ${{\cal~L}^{\rm{*}}}$.