使用線段一致性從單張影像估測三維室內曼哈頓場景

Using line consistency to estimate 3D indoor Manhattan scene layout from a single image

張星群(Hsing-Chun Chang)

指導教授 ： 賴尚宏

國立清華大學/電機資訊學院/資訊系統與應用研究所/碩士(2014年)

https://doi.org/10.6843/NTHU.2014.00636

摘要

在這篇論文中，我們提出了一套可利用單張室內場景影像估算空間布局的視覺分析演算法。我們所提出的方法，首先將室內空間視作一個三維的箱型空間，將此三維箱體投影在二維影像中，並在此立方體每一道預設平面(天花板、牆壁、地板)的投影區域中，利用所有共平面線段的一致性，以及各平面間交界線的邊界特性，合併設計出成本函數。最後以此成本函數進行旋轉、位移、箱體變型等各參數最佳化，求出準確的室內空間布局。在實驗部分，我們在成本函數中使用自適應的權重，也測試出對影像有不同先驗知識下給予權重適當調整有助於提高精確度。更進一步地，我們使用大量資料做系統穩定度測試，其中包含實際拍攝的影像和公開的室內空間影像資料庫，展示使用我們所提出的方法的準確度和穩定性。

關鍵字

三維室內場景 ； 曼哈頓世界假說 ； 線段一致性 ； 成本函數

並列摘要

In this thesis, a visual analysis approach based on computer vision algorithm is proposed to estimate the interior layout from single input image. In the proposed system, the interior space can be viewed as a three-dimensional box which includes ceiling, floor, and walls. The regions corresponding to different surfaces can be calculated by projecting the 3D box into two-dimensional image. This thesis utilizes the consistency of coplanar lines and the boundary edges between different surfaces (ceiling, floor, and walls) to design a cost function. The rotation, translation, and deformation parameters of the interior layout can be estimated by a cost function minimization process. In the experimental results, an adaptive weight parameter of the cost function is examined. The prior knowledge of real-world testing image can improve the accuracy dramatically with an intelligent weight adjustment method. Furthermore, a simulation experiment based on large-scale testing data is exploited to prove the robustness and stability of the proposed system. Finally, the proposed system also examined various real testing images, which include new collected data and the indoor scene dataset in public domain, to demonstrate the accuracy and robustness of the proposed method.

並列關鍵字

3D indoor scene ； Manhattan world assumption ； line consistency ； cost function

參考文獻

[14] A. Flint, C. Mei, I. Reid, and D. Murray. Growing semantically meaningful models for visual slam. In Conference on Computer Vision and Pattern Recognition, 2010.

[1] J. Coughlan and A. Yuille. Manhattan world: compass direction from a single image by bayesian inference. In Conference on Computer Vision and Pattern Recognition, volume 2, pages 941-947 vol.2, 1999.

[3] D. Hoiem, A. A. Efros, and M. Hebert. Recovering surface layout from an image. In International Journal of Computer Vision, 2007.

[4] Tretiak E, Barinova O, Kohli P, Lempitsky V. Geometric image parsing in man-made environments. In International Journal of Computer Vision, 2011.

[5] H. Wang, S. Gould, and D. Koller. Discriminative Learning with Latent Variables for Cluttered Indoor Scene Understanding. In European Conference on Computer Vision, 2010.