ABSTRACT
This chapter presents a research framework for progressive reconstruction of 3D building rooftops over time using multisensor data. It starts by presenting a review on building modeling, and related registration and data fusion techniques. A data-driven method is then presented to reconstruct 3D building rooftop from airborne Light Detection and Ranging (LiDAR) data by implicitly regularizing noisy data-driven geometric cues at both local and global scales using the principle of minimum description length. Next, a context-based geometric hashing method is presented to align newly acquired images with existing building models. The existing building models are finally refined by a sequential fusion approach using the Markov chain Monte Carlo method. This chapter discusses the performance of the developed methods using the International Society of Photogrammetry and Remote Sensing (ISPRS) benchmark datasets.
