|
Subject
Understanding and accurately modeling building facades from street-level imagery can significantly impact urban planning and development. By automatically detecting and analyzing architectural features such as windows, doors, and the number of floors, we can provide valuable insights into population density, traffic patterns, and resource allocation. Furthermore, this information is crucial for assessing buildings' energy efficiency and sustainability as well as for planning and executing rescue operations in disaster management.
Kind of work
In this thesis, the student will leverage advanced computer vision techniques to automatically parse building facades, detecting windows, doors, and the number of floors from panoramic images of buildings in a specific area of Brussels.
This will include: - Development of deep learning approaches, specifically focusing on window and door detection as well as floor count estimation. - Training the models using a labeled dataset, optimizing parameters to maximize detection accuracy. - Evaluating the trained models to assess their performance and conducting thorough analysis and validation to ensure they generalize well to unseen data and effectively detect architectural features across various building types and styles. The proposed approaches will be designed to be widely applicable, as street-level images are readily available across diverse geographical locations.
Framework of the Thesis
This thesis will benefit from a collaboration of ETRO with the Brussels.paradigm, which is the trusted IT partner in the Brussels-Capital Region that can be entrusted with any mission involving computer development, IT assistance, telematics, and cartography.
The student will receive datasets and initial AI and computer vision models that he/she needs to build upon.
Some relevant publications are listed below: [1] Dobson, D. (2023). Floor count from street view imagery using learning-based façade parsing. [2] He, K., Gkioxari, G., Dollár, P., & Girshick, R. (2017). Mask r-cnn. In Proceedings of the IEEE international conference on computer vision (pp. 2961-2969). [3] Li, C. K., Zhang, H. X., Liu, J. X., Zhang, Y. Q., Zou, S. C., & Fang, Y. T. (2020). Window detection in facades using heatmap fusion. Journal of Computer Science and Technology, 35, 900-912. [4] Liu, H., Xu, Y., Zhang, J., Zhu, J., Li, Y., & Hoi, S. C. (2020). DeepFacade: A deep learning approach to facade parsing with symmetric loss. IEEE Transactions on Multimedia, 22(12), 3153-3165. [5] Liu, H., Zhang, J., Zhu, J., & Hoi, S. C. (2017). Deepfacade: A deep learning approach to facade parsing. IJCAI. [6] Nordmark, N., & Ayenew, M. (2021). Window Detection In Facade Imagery: A Deep Learning Approach Using Mask R-CNN. arXiv preprint arXiv:2107.10006. [7] Sun, Y., Malihi, S., Li, H., & Maboudi, M. (2022). Deepwindows: Windows instance segmentation through an improved mask R-CNN using spatial attention and relation modules. ISPRS International Journal of Geo-Information, 11(3), 162. [8] Zhu, P., Para, W. R., Frühstück, A., Femiani, J., & Wonka, P. (2020). Large-scale architectural asset extraction from panoramic imagery. IEEE Transactions on Visualization and Computer Graphics, 28(2), 1301-1316.
Number of Students
1
Expected Student Profile
� Strong programming skills, particularly in Python, and experience with image processing and deep learning. � Motivation to explore the fields of computer vision and urban analysis.
|
|