一种基于无人机倾斜摄影的三维路网提取方法

三维路网是重要的基础地理信息数据，为了快速高效地获取包括路网在内的地表三维模型，可以采用无人机倾斜摄影技术，并设计相应的三维路网提取方法。在对无人机倾斜摄影三维模型、道路特性分析的基础上，提出一种基于无人机倾斜摄影的三维路网提取方法。首先，对无人机倾斜摄影和三维建模方法进行了分析，设计了针对三维路网提取的航空摄影策略和数据处理流程。分析了真实三维模型的构成，并结合道路在材质、形态等方面的特殊性提出了可用于道路提取的三维模型特征。然后，以三维模型中的三角面片为处理对象，开展了相关分割方法研究，消除存在混合地物的面片。利用支持向量机方法和面片的多种特征，进行了道路面片识别。最后，设计了道路面片三维连接方法和道路边缘三维修正方法，实现了面片间的连接以及道路边缘的有效修正。此外，还以北京某地区为例开展实际数据采集、三维建模和三维路网提取试验，介绍了倾斜摄影、控制点验证点数据采集以及三维模型的采集和生产过程；利用所提出的方法进行路网提取试验，得到了该区域的三维路网数据成果，并从定性和定量2个角度对成果进行评价。试验结果表明：所提出的方法可以有效地从倾斜摄影三维模型上提取路网信息，其平面和高程的精度均可满足一般导航等应用的要求。

3D Road Network Extraction Method Based on UAV Oblique Photography

Three-dimensional (3D) road network is an important basic geographical data. Unmanned aerial vehicle (UAV) oblique photography is a fast and efficient method for acquiring surface 3D models, including the road network. Based on the analysis of 3D models and road characteristics, a 3D road network extraction method based on oblique photography of UAVs is proposed. The oblique photography technique and 3D modeling methods of UAVs were analyzed, and an aerial photography strategy and data processing flow for the extraction of 3D road network were designed. The composition of the real 3D model was analyzed, and the characteristics of the road in terms of materials, morphology, etc. were presented to establish the features of the 3D model that can be used for road extraction. Taking the triangular facets in the 3D model as objects for processing, a triangular facet segmentation method was developed to eliminate the mixed ground facets. Using the support vector machine method and various features of the patch, road patch recognition was performed. A 3D connection method for road patches and 3D correction method for road edges were designed to connect the patches and for effective correction of the road edges, respectively. Using an area in Beijing as an example for data collection, 3D models were constructed and a 3D road network was extracted. The process of oblique photography, control points, and verification points for data collection, and the related 3D model were introduced. The road network extraction experiment was conducted using the proposed method, and the results were evaluated from both qualitative as well as quantitative perspectives. The experimental results indicate that the proposed method can effectively extract road network information from 3D models obtained via oblique photography, and its plane and elevation accuracy satisfy the requirements of general navigation and other applications.