Periodic Reporting for period 1 - SemanticCity (Structuring Raw Scans)
Periodo di rendicontazione: 2018-11-01 al 2020-04-30
The SmartCity: Structuring Raw Scans Proof-of-Concept project aims to develop and release a novel software suite for generating semantically structured urban models, from raw 3D scans of city neighborhoods, to streamline next-generation urban design, planning, and simulation workflows.
Several data sources now exist for large scale 3D measurements in the form of raw point clouds or polygonal soups. However, such data have to be manually cleaned and processed to produce structured output before they can be consumed by downstream applications performing engineering simulations (e.g. thermal efficiency, traffic flow). Often this (manual) step takes significantly longer than downstream simulation times resulting in a major bottleneck.
State-of-the-art workflows in design, simulation, and fabrication for urban planning (estimated market: 300+ billion euro) are now performed almost entirely in 3D. For example, buildings and large constructions are increasingly managed via building information models~(BIM) to support end-to-end digital design, fabrication, and installation. Such workflows typically have {\em stringent requirements} on the quality of input 3D models -- they are expected to be watertight, clean, and have semantic attributes. An additional challenge is to keep track of data fidelity across scale variations: building models differ from windows and vents by orders of magnitudes. However, raw 3D data as obtained from existing environments using LiDAR scans or photogrammetric images are often noisy, incomplete, corrupted with outliers, and lack any semantic organization. For example, an urban area can easily produce billions of unorganized points in a single scanning session. Significant manual cleanup and annotation is necessary before they can be consumed by downstream applications for running energy simulations (ArchSIM), previewing architectural modifications (CityEngine), or producing virtual flythrough or traffic simulation (UrbanSim). For many industrial applications, the initial data cleanup and annotation phase is the {\em major bottleneck}, both in terms of time and cost.
SmartScan addresses this challenge by automatically producing structured output directly from raw 3D measurements, and reduce computation times from weeks/days to hours. The software is based on the algorithms and technical innovations developed in the ERC-funded SmartGeometry project (StG-2013-335373), especially building on the key technical advances presented at the top conferences in our area (including a best paper and a software award). The pre-commercialization software demonstrator will be used to benchmark performance on real-world setups, to quantify the economic utility of such structured output in the context of architectural design and urban planning. As case studies, we are meaningfully utilizing large open-sourced archives of 3D measurements which are currently largely under-utilized.
Several data sources now exist for large scale 3D measurements in the form of raw point clouds or polygonal soups. However, such data have to be manually cleaned and processed to produce structured output before they can be consumed by downstream applications performing engineering simulations (e.g. thermal efficiency, traffic flow). Often this (manual) step takes significantly longer than downstream simulation times resulting in a major bottleneck.
State-of-the-art workflows in design, simulation, and fabrication for urban planning (estimated market: 300+ billion euro) are now performed almost entirely in 3D. For example, buildings and large constructions are increasingly managed via building information models~(BIM) to support end-to-end digital design, fabrication, and installation. Such workflows typically have {\em stringent requirements} on the quality of input 3D models -- they are expected to be watertight, clean, and have semantic attributes. An additional challenge is to keep track of data fidelity across scale variations: building models differ from windows and vents by orders of magnitudes. However, raw 3D data as obtained from existing environments using LiDAR scans or photogrammetric images are often noisy, incomplete, corrupted with outliers, and lack any semantic organization. For example, an urban area can easily produce billions of unorganized points in a single scanning session. Significant manual cleanup and annotation is necessary before they can be consumed by downstream applications for running energy simulations (ArchSIM), previewing architectural modifications (CityEngine), or producing virtual flythrough or traffic simulation (UrbanSim). For many industrial applications, the initial data cleanup and annotation phase is the {\em major bottleneck}, both in terms of time and cost.
SmartScan addresses this challenge by automatically producing structured output directly from raw 3D measurements, and reduce computation times from weeks/days to hours. The software is based on the algorithms and technical innovations developed in the ERC-funded SmartGeometry project (StG-2013-335373), especially building on the key technical advances presented at the top conferences in our area (including a best paper and a software award). The pre-commercialization software demonstrator will be used to benchmark performance on real-world setups, to quantify the economic utility of such structured output in the context of architectural design and urban planning. As case studies, we are meaningfully utilizing large open-sourced archives of 3D measurements which are currently largely under-utilized.