面向城市典型目标快速信息提取的车载多传感器信息融合与并行分类方法研究

At present,the vehicle-borne mobile 3D laser scanning (VMLS)technology for fast acquisition of city information is one of the central research issues in surveying and mapping. By the full integration of multisensor,abundant information of geometry and texture concerning with typical city targets such as building facades,roads, traffic signs, urban greening could be efficiently collected.Therefore,it has been playing a unique role in information extraction and reconstruction of interest objects.However,compared with the mature development of hardware, the theory of VMLS multi-source data processing for the rapid information-oriented extraction is lagging behind and can not meet the sharply increasing requirements of efficiency for various urban surveying applications.For this reason,aiming at a highly reliable classification system of urban typical targets, in this project,deep and systematical research is to be conducted on information fusion and parallel classification for vehicle-borne multi-sensor data.And the key research issues involve with the new methods and theory of massive data management,accuracy registration of panoramic image sequence and point cloud, multi-scale parallel segmentation and object-oriented classification of fusion information, followed by the development of software modules. The research is of great significance to broaden the methodology of laser scanning and to promote its applications in contemporary urban planning,geographic national conditions monitoring,etc.

面向城市信息快速采集的车载移动三维激光扫描技术是近年来国内外测绘相关领域的研究热点之一，通过多传感器的集成能够高效获取丰富的关于城市建筑立面、道路、交通标识、城市绿化等典型目标的几何与纹理信息，在目标三维信息提取与重建中有着不可或缺的独特优势。然而相对于硬件系统的成熟发展而言，其面向城市目标信息快速提取的车载多源数据处理理论与方法较为滞后，无法满足目前相关城市测绘应用的快速处理需求。为此，本项目开展车载多传感器信息融合与并行分类方法的系统深入研究，旨在于建立一套有效的针对城市典型目标的高可信度分类体系，重点研究海量点云数据管理、全景影像序列与点云高精度配准、点云多尺度并行分割以及面向对象的融合信息分类等相关的新理论与新方法，并研制软件模块。项目研究对发展和丰富三维激光扫描技术理论与方法并进一步推动其在当代城市规划与管理、地理国情监测等方面的深化应用具有重要的研究意义和应用价值。

论文开展了面向城市级车载移动三维激光扫描数据处理中的海量多源数据管理、信息融合、分割及分类等关键技术研究，是目前点云数据处理领域最具有吸引力和挑战的研究方向之一，具有重要的理论意义与研究价值。论文创新与贡献主要体现在以下四个方面：.1．提出了一种针对城市街区尺度下海量车载多源数据的管理与调度方法。在分析各类数据空间分布特性基础上，建立一套面向车载移动扫描点云、全景影像、POS轨迹、DOM等海量异构多源数据的高效管理及调度机制。.2．针对车载序列影像的匹配问题，提出了一种联合POS辅助以及KLT点特征跟踪的新匹配方法，解决序列影像中重复纹理区域的匹配问题，可高效获取分布均匀，数量充足的匹配特征点，为后期全景和点云信息的空三融合提供稳健的同名约束条件。.3．结合点云滤波与分割方法，以海量多源点云管理为基础，构建新的分类特征向量，将基于规则及SVM分类方法相结合建立了一套面向城市典型目标的层次分类体系。.4．研制了一套面向车载移动海量数据综合处理软件，提供了海量数据管理、点云滤波与分类、点云城市大比例尺测图及三维重建等模块，在多家生产及科研单位进行了试用与推广。.