近年来大场景三维重建技术在智能交通、影视娱乐、虚拟现实等领域的应用需求迅速增长，成为国内外学术界和产业界的研究热点和难点。传统的大场景三维重建方法主要基于单一的主动测量数据，如移动激光扫描点云或低空航拍影像等，由于大场景的大尺度、复杂性、动态性等特点，传统方法在数据获取成本与效率、建模精度、准确度、自动化程度等方面都存在着较大的局限性；另一方面网络平台上海量涌现的实景众包数据为场景三维重建提供了丰富的数据资源。本项目将着力研究如何将这些网络众包数据与多种方式主动采集的场景数据融合用于大场景三维重建，深入研究场景渐进式重建以及语义化重建，探索一种快速、低成本的大场景三维重建理论与方法，为现实世界三维数字化、虚拟现实仿真训练等应用提供技术支撑。

In recent years, large-scale scene reconstruction has been rapidly applied in fields such as intelligent transportation, film, television, entertainment and virtual reality etc., and been recognized as a hot, yet challenging, problem in academic and industry research. Due to the large scale, complex, and dynamic nature of the large scene, traditional approaches, based on only single active measurement data, are limited in both efficiency of data acquisition and accuracy of reconstruction. On the other hand, the massive crowdsourcing photos uploaded by users all over the world are available good resources for scene reconstruction. In this proposal, we focus on how to integrate these crowdsourcing photos with multi-source data acquired in a variety of ways for 3D reconstruction of large scenes. We propose methods for incremental scene reconstruction and semantic scene reconstruction. The research results from this project will provide fast, low-cost large-scale scene modeling methods and provide technical support for 3D digitization of real world, virtual reality and other applications.