This project proposes an automated approach to the census of technological and architectural el-ements from massive photography datasets. This use case is built on photogrammetric close-range acquisitions performed via UAV over the roofs of the centre of Bethlehem, in order to map the water tanks for civilian use that create loads on historical buildings in a seismic area. The urban census was conducted within “3D Bethlehem. Management and control of urban growth for the development of Heritage and Improvement of life in the city of Bethlehem”, a project promoted by AICS. The pre-sented work leverages the project dataset to train Deep Learning models on a Cloud Infrastructure handling model lifecycle from training to deployment. Tests were conducted on historical buildings that show, among objects of interest, multiple spurious elements such as debris and junk. Such density creates complex scenarios for models that are trained to automate recurrent operations to assist large scale monitoring and management of the areas for different teams and municipalities.
Point Cloud Segmentation for Scanto BIM: Review of Related Tecniques
The creation of as-built BIM models sees in the scan to BIM modeling one of the most time-consuming activities. Scan to BIM modeling refers to the creation of BIM objects from information derived from point clouds acquired through laser scans or photogrammetric techniques. Numerous studies have been conducted in recent years to identify automation or semi-automation procedures for the scan to BIM modeling process, which consists of different aspects: the recognition of objects within the scene, the modeling of their geometry and the recognition of the relationships between them. The present work aims to analyze actual trends in the automation of scan to BIM activities, highlighting the most used approaches and methodologies currently presented in order to provide a key to un-derstanding the development of a theme still at the dawn of its expression.
Deep Semantic Segmentation of Cultural Built Heritage Point Clouds: Current Results, Challenges and Trends
In the digital Cultural Heritage domain, the ever-increasing availability of 3D point clouds provides the opportunity to rapidly generate detailed 3D scenes to support the restoration, conservation, main-tenance and safeguarding activities of built heritage. The semantic enrichment of these point clouds could support the automatization of the scan-to-BIM processes. In this framework, the use of Artificial Intelligence techniques for the automatic recognition of architectural elements from point clouds can thus provide valuable support. The described methodology allows increasing the Level of Detail in the semantic segmentation of built heritage point clouds compared to the current state-of-the-art through deep neural networks. The main outcome is therefore the first application of DL framework for CH point clouds, with the subsequent implementation of the selected neural network (the DGCNN) for the semantic segmen-tation task. These results also permit to evaluate the pros and cons of this approach, along with future challenges and trend.
Limitations and Review of Geometric Deep Learning Algorithms for Monocular 3D Reconstruction in Architecture
This paper aims to test algorithms for 3D reconstruction from a single image specifically for building envelopes. This research shows the current limitations of these approaches when applied to classes outside of the initial distribution. We tested solutions with differentiable rendering, implicit functions, and other end–to–end geometric deep learning approaches. We recognize the importance of gener-ating a 3D reconstruction from a single image for many different industries, not only for Architecture, Engineering, and Construction (AEC) industry but also for robotics, autonomous driving, gaming, virtual and augmented reality, drone delivery, 3D authoring, improving 2D recognition and many others. Henceforth, engineers and computer scientists could benefit, not only from having the 3D representations but also from the Building Information Model (BIM) at their disposal. With further development of these algorithms it could be possible to access specific properties such as thermal, physical, maintenance, cost, and other parameters embedded in the class.
Automatic Recognition Through Deep Learning of Standard Forms in Executive Projects
In this paper is presented a possible methodology for automation through the use of deep learning of BIM modeling starting from different types of formats, such as digital processing of paper documents and CAD formats. The work is configured as a proof of concept of a possible contribution that a technique currently scarcely used in the architectural field such as deep learning can bring to the design, in particular in the realization of the information model, which today represents one of the most consuming–time activities.
From AI to H–BIM: New Interpretative Scenarios in Data Processing
The paper results from preliminary research experiences focused on the use of Artificial Intelligence as a tool for processing the large amount of data that can be obtained from digitisation processes ap-plied to the Architectural Heritage. The new interpretative scenarios will be outlined starting, on one hand, from a series of consolidated experiences in the field of three–dimensional survey, modelling and semantic enrichment, and, on the other, from the use of Augmented Reality tools for the fruition of the Heritage itself.The research aims to further investigate the great potential of processing point cloud models using Artificial Intelligence, to extrapolate, from the digitized data, information levels that go beyond shapes, offering better integration within the Building Information Modeling environment.
Deep Learning for Point Clouds Classification in the Ducal Palace at Urbino
Starting from a multi–scalar and multi–dimensional survey, most interdisciplinary researches, based on representation, are becoming a tool for dialogue between the new trends of Artificial Intelligence (AI) and the most compelling needs of our CH. The approach here proposed stems from the desire to understand how much of the skills useful in architecture analysing and modelling can be made available to the “machine”, with the goal to accelerate cognitive or management processes. Some HBIM models, as an existing digital heritage, were used to obtain the semantic intelligence. From this specialised intelligence comes a cyclical path which, through AI, transforms this knowledge into new forms of collective intelligence, at the service of the heritage. The paper presents a research that brings very promising results for the segmentation of point clouds and the facilitation of ScanToHBIM approaches, made possible by the large amount of data acquired on the Ducal Palace of Urbino.