结合2D图像和3D点云数据的城市建筑物重建关键技术研究

项目名称： 结合2D图像和3D点云数据的城市建筑物重建关键技术研究

项目编号： No.61263046

项目类型： 地区科学基金项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 储珺

作者单位： 南昌航空大学

项目金额： 45万元

中文摘要： 城市场景三维模型在智慧城市、环境感知等领域有着非常广泛的应用。具有真实感和几何准确性的建筑物重建是城市场景重建的关键技术和研究难点之一。本项目拟以街边图像为研究对象，用SFM方法获取场景的稀疏点云，建立结合2D和3D信息的高几何准确性和真实感的城市建筑重建框架。项目首先结合2D图像特征和3D信息建立高维特征向量集，通过训练多值分类器完成场景中各类物体的识别，消除行人、车辆等物体的干扰；针对消除干扰后的建筑图像提取各类物体SIFT特征及SIFT特征点结构图模型，通过机器学习方法完成建筑立面重复结构和外挂物的识别；提取2D图像中主要方向的边缘线段，结合重复结构的分布实现2D图像具有语义意义的分割；最后在2D图像分割的基础上，融合多视点图像和3D数据，完成三维点云数据的标记和优化，最终获取具有真实感和几何准确性的建筑物的三维描述和原型系统。项目的研究成果将为数字城市建设和机器人的导航提供技术支持

中文关键词： 建筑物重建；图像分割；行人检测；多视图；

英文摘要： Urban scene 3D model has a very wide application in smart city and digital entertainment, etc. Buildings modeling provided by sense of reality and accuracy geometric level is the key technique and one of difficult researches in the urban scene reconstruction. This project is to take the street view images as the research object, use sparse 3D point clounds acquired by SFM method, then combine accurate 2D and realistic 3D information to establish a urban architecture reconstruction framework. In order to finish all kinds of objects' identification and classification, we establish confluent 2D and 3D information of the high dimension characteristic matrix and trained classifier to firstly remove distraction about pedestrians and vehicles objects; With treated building image,we then extract SIFT characteristics and SIFT spatial distribution model of building image to identify the add-on objects and repetitive structure using machine learning method; According to the building's edges and the repetitive structures' distribution, complete the image segmentation with semantic meaning; Finally, accomplish the mark and the optimization of 3D point cloud datum based on the 2D intersected image and the confluent different view's 3D optimized data, to get a prototype system describing full three-dimensional reconstructed ci

英文关键词： Building reconstruction；image segmentation；pedestrian detection；multi-view image；