An Operable System for LoD3 Model Generation Using Multi-Source Data and User-Friendly Interactive Editing: Final Report

Abstract

Photo-realistic 3D city models that represent the physical and functional state of the city are necessary components of the nation’s digital infrastructure. It facilitates much easier governing, planning, simulation, measuring and prediction, and provides numerous potentials in both scientific and industry applications. LoD3 models contain building roof and façade geometry, as well as the functions of its different components (windows, doors, etc.). Generating accurate and standard 3D city models is a manually tedious, decisively rich and non-straightforward process, and the current practice of is LoD3 city modelling still a manually intensive process. Given the high demand for city-scale model production in the virtual Singapore program, we aim to develop an operable workflow that could produce LoD3 with the lightest possible manual involvement. A multi-data approach is used by integrating different sources of data including oblique imagery, aerial images, airborne/mobile LiDAR, and UAV images, to produce high quality LoD3 models that meet the CityGML standards. The workflow consists of three necessary work packages (WP) that develop techniques in 1) Geometry modelling, 2) Semantic labelling and texture mapping and 3) Integration of procedural modelling and interactive geometric editing. WP1 develops novel image-based and LiDAR based roof topography and façade geometry modelling with automated and semi-automated methods. WP2 applies data fusion techniques with the latest machine learning methods to perform land-cover classification, façade element attribution and texture mapping. WP2 also develops a preliminary proof of concept in change detection. To ensure high fidelity of the resulting models, WP3 develops novel visualization-driven editing procedures that efficiently correct errors of the models and integrate the procedural modelling workflow into the 3D reconstruction of buildings with regular geometric patterns. The major objective of this project is to work towards an operationally feasible approach to generate city-scale LoD3 models, and provide preliminary proof-of-concept on efficient model maintenance, to facilitate the broader mission of the Virtual Singapore program in developing Singapore as a more intelligent and smarter city. In the course of this project we have developed a number of novel approaches and algorithms for data processing. We have demonstrated their functionality with real data from our NUS Campus test field. However, still needed is the integration of the software packages into a fully functional operational system and a robustification of some components.