Large scale, high resolution indoor and outdoor 3d mapping with ubiquitous point clouds

点击翻译按钮获取中文摘要

The past few years have seen remarkable development in terrestrial vehicle-based mobile laser scanning (MLS) to accommodate the need for large area and high-resolution 3d data acquisition. MLS can quickly and directly acquire 3d geospatial information, which is generally identified as point clouds. As an important approach for urban scene analysis, automatic extraction of significant urban scene structures from this point cloud data has become a highly attractive research topic. The major challenges in detecting objects from large-scale laser scanning point clouds are huge data volumes, intra-class shape variation, overlap between neighboring objects, point-density variation, orientation variation, and incompleteness of object caused by occlusion. In spite of the strong interest and increased research work on MLS point cloud automatic processing, it is still an open issue. In addition, the development of smart optical sensors in urban environment provide knowledge sources complementary to MLS as well as means to propose mixed reality solution and frequently carry out change detection. However, imagery is often collected at different time with different viewpoint and resolution than MLS. Such context requires dedicated registration approaches to overcome the heterogeneous nature of the data and the differences in their acquisition configuration. Given the complexity and vast number of rapidly changing objects in urban scene, the automatic detection of urban changes is considered a major scientific challenge. By extending and advancing our on-going research for efficiently processing MLS point cloud to accommodate the need for large scale and high resolution 3d mapping, we will develop innovative techniques for extracting and labeling 3d geometric assets and for registering and matching 2d / 3d data sets available in urban context. The research program objectives include: 1) Enhancing volumetric representation and attribute-based techniques for dimensionality analysis of point clouds related to very large environments; 2) Exploring and developing techniques for the direct application and visualization of point clouds especially in browser based and mobile platform engine; 3) Developing and testing feature-based multi-resolution object recognition methods that are suitable for heterogeneous data matching; 4) Designing and testing an innovative approach for 3d MLS data matching with 2d imagery. Our grand goal for this research program is to develop and implement a comprehensive framework for large scale, high resolution indoor and outdoor 3d mapping required by emerging mobile technologies providing situated navigation and decision support solution as well as immersive and interactive mixed reality visualization. A pivotal point of this research is its focus on providing capabilities to directly visualize and exploit the point cloud without modeling the urban assets.