Advances in laser scanning technology are making 3D acquisition feasible for increasingly large objects. Different from most existing scanning projects, which have been devoted to short or medium range objects, our research focuses on dealing with long range outdoor environment scans. Compared with indoor scanning, the data obtained through outdoor scanning is large, incomplete, complex, inaccurate, and non-uniformly sampled. These properties raise many new challenges for processing outdoor scanned data. The research presented in this thesis aims to model, stylize, and render scanned outdoor environments and deliver interactive navigation.; Existing methods are normally inadequate to construct models for outdoor scanned objects such as trees and buildings because of the incompleteness and complexity of such data. In this thesis, we have developed a knowledge-based method to model outdoor scanned trees. With knowledge about the structures of trees, we use certain heuristics to produce full polygonal models from the scanned data.; The difficulty in rendering outdoor environments is accommodating the inherently incomplete and inaccurate data sets to deliver a smooth animation without suggesting the underlying data deficiency. Our solution to this issue is to employ non-photorealistic rendering techniques for rendering outdoor environment scans represented by points. In this way, varying importance and accuracy in the scanned data can be naturally illustrated. We have developed a unified framework for producing sketchy, painterly, cartoon, and intermingled styles. We have proposed a continuous level-of-detail data structure, the continuous resolution queue, to promise a coherent and consistent animation. We have also leveraged modern graphics hardware to achieve interactive rendering of large scenes.
展开▼
