About Research Project
Singapore government is highly ambitious to build a smart nation to improve the well-beings of its residents. A key component toward this vision is Virtual Singapore, a virtual city model for smart planning, organization, and management of the public resource. As a trial to the more ambitious endeavor of Virtual Singapore, Virtual NUS programme aims at building a 3D model of NUS campus to server as a test bed of key techniques for building more complex city models.
However, the current workflow of Virtual NUS is labour-intensive because it relies heavily on manual work, which is not scalable especially for a campus undergoing constant infrastructure changes. Automation or semi-automation is hardly the case in the construction of campus models and one major bottleneck is to how to classify 3D point clouds into meaningful components of 3D model.
The project aims at automating the classification of 3D point clouds of NUS campus into main buildings with roofs, walls and boundary surfaces and installations. The designed classification result is between Level of Detail 2 (LoD2) and LoD3 of CityGML (Gröger et al. 2012). Novel machine learning algorithms will be developed for the stated goal. And anotable challenge is lack of a well annotated 3D point cloud database with ground-truth labels for current application. Therefore, the outcome of this project will be twofold: 1) a 3Dpoint cloud classification benchmark facilitating 3D model reconstruction of NUS Campus and for the research community; 2) 3D point cloud classification algorithms for Virtual NUS.
