Communications of the ACM
Machine Learning Techniques Enable Models From Partial Image Data
Researchers rom King Abdullah University of Science and Technology's Visual Computing Center have developed a method to automatically complete and generate three-dimensional building models characteristic of a given area by using partial images.
Credit: King Abdullah University of Science and Technology
Researchers at Saudi Arabia's King Abdullah University of Science and Technology (KAUST) say they have integrated machine-learning techniques to develop a method to automatically complete and generate three-dimensional (3D) building models characteristic of a given area by using partial images.
"The core of our work was to come up with a meaningful set of features, parameters, and their relationships that can describe buildings generically," says Peter Wonka, a researcher at KAUST's Visual Computing Center.
Wonka notes services such as Google StreetView provide enormous volumes of data on residential buildings, and the new modeling scheme takes that data and extracts key external features of each building, such as observable footprint, size of the garage, roof style, and the window-to-wall ratio. "The scheme then 'learns' a probabilistic graphical model to encode the relationships between these features," Wonka says.
He says this enables users to sample specific features or fix observed features and compute the unobserved structure.
Finally, Wonka says there is an optimization step that translates building features into 3D building models. "Our probabilistic model for exteriors of residential buildings could also help architects more easily generate building prototypes or generate plausible 3D reconstructions from a limited set of photographs," he says.
From King Abdullah University of Science and Technology
View Full Article
No entries found