Enhancing Semantic Segmentation of Large-Scale 3D Point Clouds with Deep Learning Techniques for Urban Digital Twin Creation

ZOUHAIR BALLOUCH | 2024

Summary

Classified point clouds often serve as the primary support for decision-making scenarios. For example, these data can be used as the main layer for creating digital twins, as a basis for urban simulation studies (such as flood simulation, vegetation inventory, solar potential of rooftops, etc.), as a reference for object change detection, and as a foundation for automatic 3D modeling of the urban environment. The applications are numerous and potentially growing if classified point clouds are considered assets of digital reality. However, extracting maximum semantic information from an urban environment (parking lots, street furniture, pedestrian paths, etc.) automatically and accurately remains a challenge. Indeed, the growing development of LiDAR technology in terms of accuracy and spatial resolution offers a better opportunity to provide reliable semantic segmentation in large-scale urban environments. Thus, the development of deep learning techniques revolutionizes the field of computer vision and demonstrates high performance in semantic segmentation. The thesis clearly aims to address the challenges of accurately extracting maximum urban details from airborne LiDAR point clouds using deep learning techniques to meet the various needs of urban digital twins. We address several issues related to object extraction from airborne point clouds, particularly the adaptation of deep learning techniques, efficient fusion of point clouds with corresponding images, effective feature engineering and selection, semantic segmentation, automatic 3D modeling from semantic segmentation, visualization, and interaction with decision-making cognitive systems.

Comments

An approach to semantic segmentation of 3D point clouds developed as part of this thesis has been adopted for the implementation of projects meeting the needs of public administrations. This thesis is funded by UR SPHERES at the University of Liège, the Erasmus project, and the GeoScITY laboratory at the University of Liège.

Links

• Ballouch, Zouhair, Rafika Hajji, Florent Poux, Abderrazzaq Kharroubi, and Roland Billen. 2022. "A Prior Level Fusion Approach for the Semantic Segmentation of 3D Point Clouds Using Deep Learning" Remote Sensing 14, no. 14: 3415. https://doi.org/10.3390/rs14143415
• Ballouch, Zouhair, Rafika Hajji, Abderrazzaq Kharroubi, Florent Poux, and Roland Billen. 2024. "Investigating Prior-Level Fusion Approaches for Enriched Semantic Segmentation of Urban LiDAR Point Clouds" Remote Sensing 16, no. 2: 329. https://doi.org/10.3390/rs16020329
• Z.BALLOUCH, R.HAJJI, M.ETTARID, “The contribution of Deep Learning to the semantic segmentation of 3D point-clouds in urban areas”, IEEE International conference of Moroccan Geomatics (Morgeo), 2020.
• Zouhair Ballouch, Rafika Hajji, “Semantic Segmentation of Airborne LiDAR Data for the Development of an Urban 3D Model”, Building Information Modeling for a Smart and Sustainable Urban Space, january 2021.https://doi.org/10.1002/9781119885474.ch7
• Ballouch, Z.; Hajji, R.; Ettarid, M. Toward a Deep Learning Approach for Automatic Semantic Segmentation of 3D Lidar Point Clouds in Urban Areas. In Geospatial Intelligence: Applications and Future Trends; Barramou, F., El Brirchi, E.H., Mansouri, K., Dehbi, Y., Eds.; Springer International Publishing: Cham, Switzerland, 2022; pp. 67–77. ISBN 978-3-030-80458-9.
• Jeddoub, Imane, Zouhair Ballouch, Rafika Hajji, et Roland Billen. 2024. « Enriched Semantic 3D Point Clouds: An Alternative to 3D City Models for Digital Twin for Cities? » In Recent Advances in 3D Geoinformation Science, édité par Thomas H. Kolbe, Andreas Donaubauer, et Christof Beil, 407‑23. Lecture Notes in Geoinformation and Cartography. Cham: Springer Nature Switzerland. https://doi.org/10.1007/978-3-031-43699-4_26.