Derin öğrenme yöntemiyle semantik sınıflandırılmış nokta bulutu verisinin yapı bilgi modeli haline getirilmesi
Creation of bim model from semantically segmented point cloud via deep learning
- Tez No: 800328
- Danışmanlar: PROF. DR. HANDE DEMİREL
- Tez Türü: Yüksek Lisans
- Konular: Jeodezi ve Fotogrametri, Mühendislik Bilimleri, İnşaat Mühendisliği, Geodesy and Photogrammetry, Engineering Sciences, Civil Engineering
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 2023
- Dil: Türkçe
- Üniversite: İstanbul Teknik Üniversitesi
- Enstitü: Lisansüstü Eğitim Enstitüsü
- Ana Bilim Dalı: Geomatik Mühendisliği Ana Bilim Dalı
- Bilim Dalı: Geomatik Mühendisliği Bilim Dalı
- Sayfa Sayısı: 63
Özet
Mimarlık, Mühendislik ve İnşaat sektörü açısından yapı bilgi modellemesinin uygulanması maliyet, dijitalleşme ve sürdürülebilirlik açısından önemli bir yer tutmaktadır. Yapı bilgi modellemesi, sağladığı 3 boyutlu geometrik ve semantik veri ile binaların yaşam döngüsünün etkili bir şekilde yürütülebilmesine olanak sağlamaktadır. Yapı bilgi modellinin oluşturulması bu sürecin önemli bir kısmını kapsamaktadır. Günümüzde yapı bilgi modeli 2 boyutlu çizimler veya 3 boyutlu nokta bulutu kullanılarak manuel olarak oluşturulabilmektedir. Bu yöntemler kullanışlı olsa da manuel olmalarından dolayı zaman kaybı yaratmaktadırlar. Bu zaman kaybı gerekli otomasyon sağlanarak minimize edilebilmektedir ancak bu konuda günümüzde geniş çaplı kullanılan bir yöntem bulunmamaktadır. 2 boyutlu CAD çizimleri bu otomasyonun sağlanmasında sadece yatay konum bilgisi sağladıkları için çok verimli olmamaktadır. Nokta bulutu verisi ise sağladığı yüksek detaylı 3 boyutlu geometrik veri ile otomatik modelleme süreçleri için daha uygundur. Gelişen ölçüm teknolojileri ile birlikte mekânsal veri elde etmek için kullanılan klasik yöntemler yerini lazer tarama ve insansız hava aracı yardımıyla ölçüm gibi daha modernize edilmiş yöntemlere bırakmışlardır. Özellikle lazer tarama yöntemi sağladığı 3 boyutlu geometrik veri ile Yapı bilgi modelinin oluşturulmasında önemli bir rol oynamaktadır. Lazer tarama sonucu elde edilen nokta bulutu verisi geometrik özellikleri yüksek doğrulukta yansıtmasına rağmen bina içi objelerin sınıflandırılması konusunda herhangi bir semantik veri sağlamamaktadır. Bu sebeple nokta bulutu verisinin çeşitli yöntemler ile otomatik sınıflandırılması sağlanmalıdır. Bu tezin kapsamında lazer tarama ile Gizil Enerji ofisinindeki bir bölümün nokta bulutu verisi elde edilmiş ve bu elde edilmiş nokta bulutunun derin öğrenme yöntemi ile sınıflandırılması sonucunda elde edilen veri kullanılarak otomatik şekilde yapı bilgi modelinin oluşturulması amaçlanmıştır. Bu kapsamda tarama sonucu elde edilen nokta bulutu Pointnet algoritması kullanılarak bina içinde modellemeye konu olacak tüm objeler sınıflandırılmış, sınıflandırılan bu veri K-Mean algoritması, RANSAC algoritması algoritması kullanılarak kendi içinde sınıflara ayrılmıştır. En son aşamada ise Autodesk Revit programı içerisinde yer alan Dynamo görsel programlama yazılımı kullanılarak sınıflandırılan nokta bulutu verisinden yapının geometrik ve semantik özelliklerini yansıtan BIM modeli oluşturulmuştur.
Özet (Çeviri)
In the last few decades, the world has been facing significant challenges that threaten human survival. One of these challenges is resource scarcity, which is mainly caused by overpopulation, environmental degradation, and unsustainable consumption patterns. The high population rates have led to the increased demand for buildings, which has put pressure on natural resources such as wood, water, and energy. The construction and operation of buildings account for a significant portion of energy consumption and greenhouse gas emissions, which contributes to climate change and environmental degradation. The Architecture, Engineering, and Construction (AEC) industry has a critical role to play in addressing these challenges by designing and constructing sustainable buildings that minimize resource consumption, reduce waste, and enhance the well-being of building occupants. Sustainable buildings are designed to minimize their environmental impact throughout their life cycle, from construction to operation and demolition. This requires a holistic approach that integrates environmental, social, and economic considerations into the design and construction process. Building Information Modelling is a pioneering approach for building sustainable structures for better environment. Building Information Modelling (BIM) is a digital representation of physical and functional characteristics of a building. BIM provides a comprehensive view of the building design, construction, and operation, which enables stakeholders to make informed decisions and optimize building performance throughout its life cycle. BIM is a powerful tool that supports sustainable building design and operation by providing accurate and timely information about the building's environmental impact, energy performance, and occupant comfort. BIM enables stakeholders to minimize or eliminate negative impacts on the environment and society by providing a platform for collaborative decision-making, design optimization, and resource management. BIM can help reduce construction waste, improve energy efficiency, and enhance building occupant comfort and well-being. BIM is increasingly being adopted by the AEC industry as a standard practice for building design and construction. However, BIM is mostly used for new construction projects, and there are significant challenges in creating BIM models for existing buildings. Existing buildings account for a significant portion of global building stock, and they present unique challenges for sustainable building design and operation. Many existing buildings were built before the advent of BIM, and their design and construction information may not be available in a digital format. Therefore, creating BIM models for existing buildings is essential for optimizing their performance and minimizing their environmental impact. Creating BIM models for existing buildings involves converting physical building elements into a digital format that can be used for design and construction purposes. There are different ways to create BIM models, such as creating BIM models from 2D CAD drawings, 3D objects, and point clouds. Creating BIM models from 2D CAD drawings is a common approach that involves converting 2D drawings into 3D models. This method is relatively simple and requires less processing power compared to other methods. However, this method may not provide an accurate representation of the building's geometry and functionality, especially for complex buildings. Creating BIM models from 3D objects is another approach that involves using 3D modeling software to create the BIM model. This method provides a high level of accuracy and detail, and it can be used to represent complex geometries and functionalities. However, this method requires significant expertise in 3D modeling, and it can be time-consuming and expensive. Creating BIM models from point clouds is a promising approach for existing building documentation because it provides a highly accurate and detailed representation of the building's geometry. However, creating BIM models from point clouds presents significant challenges, such as data complexity, data processing, and data interpretation. Therefore, the use of advanced technologies such as deep learning and semantic segmentation is essential for creating accurate and efficient BIM models from point clouds. The creation of BIM models from point clouds can be achieved through different methods, such as manual modeling, semi-automatic modeling, and creation from semantically segmented point clouds. Each method has its advantages and drawbacks. Manual modeling involves creating a BIM model by manually modeling each object and surface in the point cloud. The advantage of manual BIM creation from point clouds is that it allows the modeler to have complete control over the modeling process and can be more flexible and adaptable than other methods. For example, the modeler can easily adjust the level of detail or complexity of the model as needed to meet specific project requirements. This method provides a high level of accuracy and detail, but it can be time-consuming and expensive, especially for complex buildings. Semi-automatic modeling involves using software tools to assist in creating the BIM model, such as automated feature extraction and object recognition. The software uses algorithms and machine learning techniques to automatically identify and extract features such as walls, doors, and windows from the point cloud data. The modeler can then review and edit the extracted features as needed to ensure accuracy and completeness. This method provides a more efficient way of creating BIM models, but it may not provide the same level of accuracy and detail as manual modeling. Creation from semantically segmented point clouds involves labeling point clouds with semantic information, such as object types, surface materials, and functional elements, and using deep learning algorithms to create the BIM model. This method provides a structured and meaningful representation of the building's geometry and functionality, which enables stakeholders to make informed decisions and optimize building performance. This method is also more efficient compared to manual modeling and provides a higher level of accuracy and detail compared to semi-automatic modeling. Especially, within the developments in deep learning technologies, semantic segmentation of point cloud become a significant topic for researchers. To sum up, The advantage of using deep learning techniques is that they can significantly reduce the time and effort required to create a BIM model of an existing building. Additionally, they can improve the accuracy and completeness of the model by identifying and segmenting elements that might be missed by manual or semi-automatic methods. The scope of this thesis is creating a BIM model from semantically segmented point cloud data automatically via deep learning approach. Firstly, point cloud data of Gizil Energy office was acquired by laser scanning,then raw point cloud has been registered in Leica's Cyclone Software. The obtained point cloud was classified by using PointNet which is a deep learning algorithm. Then, these semantically classified data were categorized into classes using K-Mean and RANSAC algorithms. In the final stage, the BIM model representing the geometric and semantic features of the structure was created from the classified point cloud data in Dynamo software, which is a component of Autodesk Revit Software.
Benzer Tezler
- Deep learning-based techniques for 3D point cloud analysis
3B nokta bulutu analizi için derin öğrenme temelli teknikler
YUSUF HÜSEYİN ŞAHİN
Doktora
İngilizce
2023
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrolİstanbul Teknik ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
PROF. DR. GÖZDE ÜNAL
- GPT modelleri kullanılarak ironi tespiti ve sınıflandırması
Irony detection and classification using GPT models
MUSTAFA ULVİ AYTEKİN
Yüksek Lisans
Türkçe
2024
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolGazi ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
PROF. DR. OSMAN AYHAN ERDEM
- Specific keyword extraction from unstructured curriculum vitae using deep learning methods
Derin öğrenme yöntemleri kullanılarak özgeçmişler üzerinde anahtar kelime çıkarımı
MUSTAFA BUĞRA DÜR
Yüksek Lisans
İngilizce
2021
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolÇankaya ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
PROF. DR. HAYRİ SEVER
- Automatic reconstruction and efficient visualization of 3D city models
3B şehir modellerinin otomatik üretimi ve etkili görselleştirilmesi
MEHMET BÜYÜKDEMİRCİOĞLU
Doktora
İngilizce
2023
Jeodezi ve FotogrametriHacettepe ÜniversitesiGeomatik Mühendisliği Ana Bilim Dalı
PROF. DR. SULTAN KOCAMAN
- Özellik tabanlı görüş madenciliğinde yapay zeka teknikleri kullanarak görüş hedefi çıkarımı ve kategori tespiti
Opinion target extraction and category detection using artificial intelligence techniques in aspect-based opinion mining
KÜRŞAT MUSTAFA KARAOĞLAN
Doktora
Türkçe
2022
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolKarabük ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
PROF. DR. OĞUZ FINDIK