The housing market is influenced by various factors that impact the pricing of properties. One such factor is the internal layout and structure of the estate, which is typically represented by a floor plan. Real estate agents commonly use floor plans to provide buyers with a comprehensive overview of the property. This thesis aims to extract information from floor plans and investigate if certain structural factors correlate with the final price of an estate.

To extract information from floor plan images, computer vision tools are employed, specifically convolutional neural networks (CNNs). The chosen model for this study is Mask R-CNN, a CNN-based model capable of performing instance segmentation on a dataset of images. The segmentation results are then visualized and interpreted to identify key structural features.

The findings of this research demonstrate that instance segmentation can be achieved with a reasonable level of precision. Furthermore, there is a discernible tendency indicating a potential pattern between the extracted information and the final price of the estate. However, the chosen parameters for this thesis do not exhibit a significant correlation. To establish a stronger correlation, further in-depth and meticulous studies need to be conducted.

Overall, this thesis contributes to the understanding of how structural factors in floor plans may influence property prices. The application of deep learning techniques, specifically Mask R- CNN, showcases the potential of computer vision in extracting valuable insights from real estate floor plans. The limitations and future directions highlighted in this study provide a basis for further research in this domain.