@phdthesis{,
    author = {Rahmani, Kujtim},
    title = {Facade Interpretation from Images},
    editor = {},
    booktitle = {},
    series = {},
    journal = {},
    address = {},
    publisher = {},
    edition = {},
    year = {2020},
    isbn = {},
    volume = {},
    number =  {},
    pages = {},
    url = {},
    doi = {},
    keywords = {Computer vision, Facade Segmentation, Structured Random Forest, Object Detection, Transoms and Mullions},
    abstract = {Artificial Intelligence (AI) has reached every area in our life. Furthermore, three-dimensional (3D) models are becoming a main topic of technology. This thesis is devoted to using AI techniques for facade segmentation which aids the 3D reconstruction of the facades of buildings. First, we have developed a pipeline for the segmentation of whole facade distinguishing the classes balcony, door, roof, shop, sky, wall and window. Our hybrid pipeline consists of traditional and modern machine learning approaches. The Structured Random Forest is based on feature extraction, our object detector employs a deep learning approach and the final part model fitting uses dynamic programming. The pipeline is evaluated on several datasets, it performs better than or on a par with currently published methods and it is the state of the art for small datasets. Second, we have built a system for window delineation segmenting every window in window frame, mullions and transoms, giving us a complete description of a window. Similarly to the previous one, the system uses a pipeline consisting of an object detector, semantic segmentation employing deep learning and model fitting based on the geometric shape and appearance of the windows. For semantic segmentation an architecture is chosen, which can capture thin elements in an image. Additionally, also dynamic programming and information fusion are used to create the final segmentation result. We are the first that have built a system for the segmentation of window frame, transoms and mullions as well as a high-resolution dataset for training and evaluation of the detector for transoms and mullions. We have evaluated the relevance of features and the optimization functions of the nodes in the decision trees. Additionally, we have shown experimentally the effect of feature cleaning on the final results as well as what features are important. Finally, we have demonstrated that we have built a reliable system for facade segmentation which can be used as a source for building high-quality 3D building models.},
    note = {},
    
    school = {Universität der Bundeswehr München},
    
}