中文摘要：

基于单幅建筑物图像,提出一种以简单交互方式累进式重建三维建筑物场景的方法.在以单幅建筑物图像为背景的画布上,用户首先标记建筑物的2个基面并对输入图像进行校正预处理,校正过程中充分利用透视图像中的平行线交于同一个图像消影点的特点,由射影变换快速消除建筑物图像中的透视失真并恢复建筑物图像中的线条平行关系;然后在经校正处理后的图像上描绘建筑物基本轮廓线条,系统自动查找轮廓线条平行垂直关系,利用二维线画图中各线条的平行、垂直及顶点共面关系作为约束建立能量函数计算线画图中所有顶点的深度值,并根据结构对称性特点完善建筑物模型几何结构,生成建筑物场景的粗糙模型;最后利用建筑物图像呈现的精细细节结构和其与建筑物粗糙模型主体结构的几何位置关系确定细节结构顶点的深度值,并根据输入图像中可见部分纹理信息进一步合成建筑物表面纹理,得到精细真实的三维建筑物模型.实验结果表明,该方法能够方便、高效地重建输入图像中的建筑物场景,并且具有较强的真实感.

英文摘要：

Based on the input single architecture image, a novel 3D reconstruction approach for urban architecture is presented in this paper, which can create real-world buildings in an interactively progressive manner. Upon the sketching plane defined by the input image, the users firstly mark two main faces of the target architecture for image rectification. Due to the fact that the parallel lines under perspective projection always intersect at a vanishing point, the perspective distortions can be removed and the parallel relations among sketch lines can be restored by a special projection transformation. Then, the users interactively sketch the visible primary profiles of target object on the rectified image, and the system automatically identifies different kinds of parallel or vertical relations of the sketch lines. All of these relations and co-plane properties can be regarded as the geometric constrains for calculating the z-depths of sketch lines. Owing to the property of symmetry, the coarse model can thus be generated by completing the geometric structures of architecture scene. Furthermore, by incorporating the geometrical relations between urban details and the primary structure into the z-depth computation of architectural details, a real fine 3D architecture can finally be reconstructed by synthesizing the surface texture derived from visible texture information of the input image. Experimental results show that our proposed method can conveniently and effectively reconstruct 3D architecture model from single image with high visual quality.