Procedural 3D Reconstruction and Quality Evaluation of Indoor Models
Total Page:16
File Type:pdf, Size:1020Kb
UNIVERSITY OF MELBOURNE Melbourne School of Engineering Procedural 3D Reconstruction and Quality Evaluation of Indoor Models by Ha Thi Thu Tran A thesis submitted in total fulfilment of the requirements of the degree of Doctor of Philosophy in the Department of Infrastructure Engineering September 2019 Declaration of Authorship This is to certify that 1. the thesis comprises only my original work towards the PhD; 2. due acknowledgement has been made in the text to all other material used; 3. the thesis is less than 100,000 words in length, exclusive of tables, maps, bibliographies and appendices. Signed: Ha Thi Thu Tran Date: September 10, 2019 i Abstract Building Information Modelling (BIM) plays an important role in the digital transformation of the construction sector and built environments. BIM promises to achieve better quality infrastructures and to shorten the duration of construction projects, and also has additional values in the global infrastructure market, as it potentially provides more efficiency in collaboration, transparency and information management, and greater intelligence in the decision-making process during the whole lifetime of buildings. In addition, up-to-date 3D building models serve as a versatile data source for various applications such as energy simulation, navigation, location-based services, and emergency response. Today, 3D building models are available only for newly designed or recently constructed buildings. A large proportion of existing buildings have been in existence for many decades. Meanwhile, the 3D models are not often updated to reflect changes in an existing building during the different stages of its lifecycle. Automated methods for efficient and reliable generation of 3D building models have the potential to expand the application domains to existing buildings. Although recent lidar scanning and photogrammetry techniques allow efficient capturing of the as-is condition of the built environment, the development of automated processes for the production of accurate, correct, and complete 3D models from the data remains a challenge. In addition, quantitative measurement of the quality of the 3D reconstructed models is essential, as it enables the measurement of the faithfulness of the models in representing the physical built environment. This thesis aims to develop a novel approach to automated reconstruction of indoor models and to provide a comprehensive method for quantitative evaluation of the quality and change detection of indoor models. The thesis contains three main contributions. First, a shape grammar approach for procedural modelling of indoor environment containing Manhattan world designs from lidar data is proposed. The hypothesis of this research is that understanding and translating the principles of indoor architectural design into a modelling algorithm will provide the capability to overcome the challenges and assist the reconstruction of a 3D semantic-rich model. Second, a procedural method for automated reconstruction of generic ii indoor models (i.e., Manhattan and non-Manhattan world buildings) using a stochastic approach is developed. The approach is based on a combination of a shape grammar and a data-driven approach, which facilitates the automated application of grammar rules in the production process and enhances its robustness to incomplete and inaccurate input. Third, a comprehensive method for quality evaluation, comparison, and change detection of 3D indoor models is proposed. The evaluation method facilitates a quantitative assessment of geometric quality of indoor models in terms of three quality aspects: completeness, correctness, and accuracy. The change detection method enables identification of redundant elements in existing 3D models and the missing elements in indoor environments. A series of experiments was carried out to evaluate the performance of the proposed methods on synthetic and real datasets, and the results show the capability of the methods for the reconstruction of complex indoor environments with high accuracy, completeness, and correctness. iii Abbreviations BIM Building Information Modelling 2D Two Dimensional 3D Three Dimensional MW Manhattan World Non-MW Non-Manhattan World IFC Industry Foundation Classes GIS Geographic Information System RGB-D Red Green Blue - Depth OGC Open Geospatial Consortium CityGML City Geography Markup Language IndoorGML Indoor Geography Markup Language LOD Level of Detail B-rep Boundary representation CSG Constructive Solid Geometry TLS Terrestrial laser scanner MLS Mobile laser scanning CGA Computer Generated Architecture SfM Structure from Motion SLAM Simultaneous Localisation and Mapping iv L-system Lindenmayer systems rjMCMC reversible jump Markov Chain Monte Carlo v Publications List of the peer-reviewed publications: Refereed Journal Publications Tran, H., Khoshelham, K. and Kealy, A., 2019. Geometric comparison and quality evaluation of 3D models of indoor environments. ISPRS Journal of Photogrammetry and Remote Sensing, 149, pp.29-39. Tran, H., Khoshelham, K., Kealy, A. and Díaz-Vilariño, L., 2018. Shape Grammar Approach to 3D Modelling of Indoor Environments Using Point Clouds. Journal of Computing in Civil Engineering, 33(1), p.04018055. Submitted Journal Publications Tran, H., Khoshelham, 2019. Procedural reconstruction of 3D indoor models from lidar data using reversible jump Markov Chain Monte Carlo. ISPRS Journal of Photogrammetry and Remote Sensing. Peer-reviewed Conference Publications Tran, H. and Khoshelham, K., 2019. A stochastic approach to automated reconstruction of 3D Models of interior spaces from point clouds. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences (Best paper award – Indoor3D workshop at ISPRS Geospatial Week 2019). Tran, H., Khoshelham, K., Kealy, A. and Díaz-Vilariño, L., 2017. Extracting topological relations between indoor spaces from point clouds. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 4, p.401. Tran, H. and Khoshelham, K., 2019. Building change detection through comparison of a lidar scan with a building information model. International Archives of the Photogrammetry, Remote Sensing & Spatial Information Sciences (ISPRS Geospatial Week 2019). vi Khoshelham, K., Tran, H., Acharya, D., 2019, Indoor mapping eyewear: Geometric evaluation of spatial mapping capability of Hololens. International Archives of the Photogrammetry, Remote Sensing & Spatial Information Sciences (ISPRS Geospatial Week 2019). Khoshelham, K., Tran, H., Díaz-Vilariño, L., Peter, M., Kang, Z. and Acharya, D., 2018. An evaluation framework for benchmark indoor modelling methods. International Archives of the Photogrammetry, Remote Sensing & Spatial Information Sciences, 42(4). vii Acknowledgements I would like first to express my grateful appreciation of my principal supervisor, Dr. Kourosh Khoshelham. I thank him deeply for the excellent supervision, great supports and encouragement that he provided throughout the years of my doctoral studies. I would like to thank him not only for helping me to gain skills and enrich my knowledge of the field, but also for constructively challenging my initial ideas whilst at the same time allowing me the freedom to shape those ideas. His relentless guidance, patience, and kindness inspire me to develop a deeper interest and stronger motivation for doing research and other academic activities. It has been truly a great experience and I feel very fortunate to have worked under his supervision. I would also like to thank my co-supervisor Professor Allison Kealy and the chair of my advisory committee Professor Stephan Winter for their support, suggestions, and encouragement. Special thanks go to Professor Stephan Winter for helping me to find my principal supervisor. Without his help and support, I might not have had the chance to do my doctoral studies at the University of Melbourne. I would like to extend my acknowledgement to Professor Tuan Ngo, who gave me the opportunity to join an industrial project, which provided me with better understanding and enables me to see the potential of applying my knowledge in the construction sector. I also thank my mentors, Dr. Noel Faux, Dr. Behzad Bozorgtabar and Dr. Suman Sedai for their guidance, support and sharing of industrial experience during my three-months internship at IBM research–Australia. I also want to extend my deep gratitude to all my lab mates, Milad Ramezani, Fuqiang Gu, Debaditya Acharya, Yan Li (IE), Yan Li (EE), Hao Chen and Hanxian He, not only for fruitful discussion and exchange of useful information, but also for their encouragement and cheerful moments. viii I feel very lucky to have friends who care about me and are always there for me when I need them. I would like to thank my close friends Dr. Nguyen Xuan Thu, Mr. John Drennan, Dr. Thu Phan, Mr. Sengor Kusturica, Dr. Ken – Ho Le Khoa, Dr. Kim Quy, and Dr. Kim Anh Thi Dang. Their friendship has been a great support during my studies, and they mean a lot to me in my life. I specially thank Dr. Thu Phan, who encouraged me to pursue my doctoral study and provided great support during my application. I also particularly thank Mr. John Drennan for his great help with my writing and for proofreading my thesis. Finally, I wanted to express my love and appreciation of my parents, Mr. Tran Van Dung and Mrs. Tran Thi Vy, and my sister, Dr. Tran Thi Thanh Huyen. Their unconditional love and support are always the greatest inspiration and motivation in my life. I sincerely