Abstract: Recently, the 3D reconstruction of indoor scene becomes a hot spot of research, as the popular of virtual house. Based on image understanding, this paper proposes an image-based box modeling method of 3D indoor scene reconstruction. Firstly, the vanishing points of indoor scene image are detected to represent the main axis of the room. With the analysis of algorithm??s influence factors, this paper improves the vanishing points detection algorithm to       ensure efficiency and accuracy. Then, this paper recovers automatically the scene spatial layout information with the layout reasoning steps which significantly improve the result. Finally, the calculation of camera parameters is       performed by using vanishing point, so the conversion from image scene to 3D model is implemented. The experiments demonstrate that the proposed method can quickly analyze the indoor scene images and detect the 3D layout for virtual scene reconstruction.

Key words: virtual house, image understanding, vanishing points detection, 3D reconstruction, box modeling of   , indoor scene

摘要： 近年来，随着家居虚拟展示应用的推广，针对图像的室内场景建模技术成为研究和应用的热点。在图像理解的基础上，针对单张复杂室内图像提出了一套箱体建模方法。首先进行代表房间主轴方向的灭点检测，并通过分析影响灭点检测的因素，对灭点检测算法提出相应的改进，从而显著改善算法的执行效率和准确性。然后对图像的场景布局信息进行自动还原，并通过加入布局优化步骤，提高布局还原结果的准确率。最后利用灭点得到相机的内、外部参数，并以此为基础实现图像到三维模型的转换。实验表明，单张复杂室内图像的箱体重建方法能够快速地对图像场景进行分析，并恢复场景的箱体布局，满足虚拟重建的需要。

关键词: 家居虚拟展示, 图像理解, 灭点检测, 三维重建, 室内场景箱体建模