基于结构先验与协同优化的城市场景分段平面重建

Piecewise Planar Urban Scene Reconstruction Using Structure Priors and Cooperative Optimization

在基于图像的城市场景三维重建中,场景分段平面重建算法可以克服场景中的弱纹理、光照变化等因素的影响而快速恢复场景完整的近似结构.然而,在初始空间点较为稀疏、候选平面集不完备、图像过分割质量较低等问题存在时,可靠性往往较低.为了解决此问题,本文根据城市场景的结构特征构造了一种新颖的融合场景结构先验、空间点可见性与颜色相似性的平面可靠性度量,然后采用图像区域与相应平面协同优化的方式对场景结构进行了推断.实验结果表明,本文算法利用稀疏空间点即可有效重建出完整的场景结构,整体上具有较高的精度与效率.

In the image-based 3 D scene reconstruction, piecewise planar stereo methods can effectively recover complete and approximate structures of a scene by overcoming some interference factors such as textureless regions and illumination variations. However, they could be unreliable when some problem(e.g., sparse initial 3 D points, incomplete candidate planes and inaccurate image over-segmentation) is involved. In order to address these problems, according to the structure characteristics of urban scenes, a novel plane reliability measurement is presented by incorporating scene structure priors,3 D point visibility and color similarity. Then, the scene is reconstructed in a progressive refinement manner so as to jontly optimize image regions and their associated planes. Experimental results on a variety of urban scenes show that the proposed method can effectively reconstruct the complete structures of a scene with high efficiency and accuracy only using sparse 3 D points.