Three Dimensional Street Scene Representation Learning for Street Frontage Classification

Street View Imagery (SVI) serves as a valuable medium for sensing and analysing street spaces, however, it inadequately represents the 3D morphological structure of streets. The study presents a novel approach that transforms a single omni-directional SVI into a coloured point cloud, explicitly enhancing its ability to represent the 3D morphological structure of street space. Building on these point clouds, experiments are conducted applying 3D deep learning methods to infer street frontage attributes, such as the activeness. Street frontage classification based on 3D point cloud structures performs comparably to traditional image-based methods and surpasses them with the addition of both 3D and colour information. The work presents great potential to understand and explain street attributes from street morphology, contributing to more effective urban design and renewal strategies.

Recommended citation: Fan, Z., Law, S., Biljecki, F. (2025). Three Dimensional Street Scene Representation Learning for Street Frontage Classification. In The 19th International Conference on Computational Urban Planning and Urban Management (CUPUM), London, UK.
Download Paper