Data-Driven Procedural Modeling on Interiors

We propose new methods to allow efficient and interactive procedural generation of building interiors. The goal will be to use procedural modeling, graph theory and optimization methods for the automated design of highly detailed, furnished rooms. Our techniques will greatly extend the range of high-quality indoor environments that can be efficiently created by individual users. This will be achieved by (1) a high level of abstraction by description of object classes, (2) the automated creation of designs, and (3) the reusability and variability of created designs. Target applications include indoor architecture design, everyday furniture planning, real estates marketing, digital content creation for games and movies, facility management, historical and archaeological visualizations and many more. For automated model creation, we will improve the concept of shape grammars. A data-driven design approach will transform semantic description of interior designs to geometric models. It will select and combine grammar rules from a special evolving repository and use the resulting procedures for creation of a large variety of original models. User interaction will be minimized by utilization of abstract descriptions, however direct geometry editing will be still possible. Our approach will also allow efficient editing of existing designs with automatic preservation of semantic and geometric plausibility. As interiors are often created in many iterations, the costs of refinement steps and error corrections will be strongly reduced. Current procedural techniques for interiors are focused on very specific, partial problems. Our results will become a part of a unified content production pipeline for creation of multi-scale urban environments from the city level all the way to furnished interiors of single houses. The research will be divided into three main computer graphics basic research problems: 1. Floor planning divides a floor into rooms. Semantics and architectural knowledge determine their connections, sizes and shapes. Our contribution will be a graph-based layout algorithm for arbitrarily shaped rooms with a guarantee of no empty spaces and a minimum of parameters. 2. Furniture placement arranges furniture items inside of a room. Mutual relations, functionality, ergonomics and room style determine the layout. Our contribution will be an automated optimization of layouts for illumination, ergonomics and emergency situations, as well as high-level control of layouts by semantic styling. 3. Furniture generation creates furniture models, manages the configuration of their movable parts and applies textures. Our contribution will be general polyhedral furniture shapes, model generation inside of constrained spaces, and kinematics of models and posture selection.

The main goal of our research work was to develop efficient, interactive methods for generation of 3D models of man-made objects.Innovations in digital technologies push forward the quality of virtual environments we daily encounter on the screens of our mobile devices, desktops and TVs. The amount of details, scale and realism of 3D content for films, computer games, advertising, urban planning, real estates marketing, and for historical and architectural visualizations is required in constantly increasing volumes and quality. New VR and AR devices, as well as UHD screens, increase the demand even more.The 3D content required for these applications is being produced by artists, designers and architects. Designing mostly by hand, their capacities are limited. Our research was focused on computer-graphics methods for efficient, interactive generation of 3D objects and sceneries. The results of our project allow high-quality environments to be efficiently created by small production teams or even by individuals. Design costs are reduced because our methods: automatically adapt, reuse and vary existing content, allow high-level input, e.g., using keywords instead of direct geometry manipulation, and learn to combine partial designs and predict their quality. We conducted basic research on various topics related to creation of virtual buildings and their interiors. We analyzed how façade elements can be divided into layers, each following a simple pattern. Patterns allow estimating the façade structure on photographs, where parts of the building are hidden by obstacles like trees, vehicles, etc. Our next tool assists with interactive floor planning by identifying structurally sound designs. It seeks the best configuration of supporting columns to carry the load of the mass of the floors above. Further, we created a framework for mapping furniture arrangements between various room shapes. The most important result of this project is a general system for the generation of buildings, which is the first to be focused on collaborative modeling. It observes how its users work and learns to assemble design pieces automatically to large 3D scenes.