基于语义约束和多源数据的城市场景三维重建方法

3D reconstruction of urban scenes is an important research topic in the field of computer graphics, computer vision and geographic information science, with widely applications in smart city and virtual reality. However, due to the complexity of objects in urban scenes and limitation of data acquisition devices, existing methods still lack the ability of reconstructing the 3D representation of urban scenes efficiently and accurately. To solve these problems, this project aims to develop novel 3D reconstruction methods of urban scenes based on semantic constraints and multi-source data. The proposed methods utilize high-level semantic features to deal with the ambiguity caused by noise and data missing, thus overcome the deficiency of traditional geometric feature based methods which is sensitive to the quality of original point cloud data. The project will cover three major aspects: semantic information mining of urban scenes, feature fusion of multi-source data and geometric modeling based on semantic constraints. Several key scientific problems will be tackled, including semantic feature extraction of urban scenes, consistency control of feature fusion from multi-source data and description of semantic constraints in geometric space. Technical innovations of the project include multi-dimension and multi-scale semantic feature extraction of urban scenes, multi-source data feature fusion based on probabilistic graph model and semantic constraints based mesh generation.

城市场景的三维重建是计算机图形学、计算机视觉以及地理信息学科中的热点问题，在智慧城市、虚拟仿真等领域具有重要的应用价值。由于城市场景包含对象的复杂性以及数据采集设备的局限性，现有的三维重建技术难以满足对城市场景进行高效准确建模的要求。本项目针对这一问题，研究新的基于语义约束和多源数据的城市场景三维重建方法，从而有效克服传统的基于底层几何特征的研究方法过于依赖原始点云质量的不足，利用高层语义特征有效解决噪声和数据缺失导致的二义性问题。主要研究内容包括城市场景语义信息挖掘、多源数据信息的特征融合以及基于语义约束的几何建模。关键科学问题包括城市场景语义化特征提取问题、多源数据特征融合中的一致性处理问题以及网格重建中语义约束的几何空间描述问题。主要技术创新包括城市场景语义特征的多尺度和多维度提取、基于概率图模型的多源数据特征融合以及基于语义约束的网格重建。

城市场景的三维重建是计算机图形学、计算机视觉以及地理信息学科中的热点问题，在智慧城市、虚拟仿真等领域具有重要的应用价值。由于城市场景包含对象的复杂性以及数据采集设备的局限性，现有的三维重建技术难以满足对城市场景进行高效准确建模的要求。..为实现既定的研究目标，开展了3个方向的研究工作：城市场景语义信息挖掘、多源数据信息的特征融合以及基于语义约束的几何建模。通过研究，建立了多源城市场景数据语义目标提取的基础理论和方法。共发表高水平SCI论文7篇，国际会议论文1篇，申请发明专利2项。