Classification of Photo-Realistic 3D Window Views in a High-Density City: The Case of Hong Kong

Window view is an intimate medium between occupants and nature, especially in high-density cities like Hong Kong; and thus belongs to the quality of a house or apartment. In literature, researchers found that window views of nature are vital to the occupants’ physical and psychological health and productivity improvement. Understanding the view situation at the urban level can facilitate urban environment optimization, urban planning and development policies, and smart city management. Currently, views of nature have been quantitatively studied in satellite images and cars’ cameras at a macro or micro level, respectively. However, as an essential supplement to the greenery view information hub at a mesoscale, few studies on efficient visualization and classification of window views at the urban level seem available. This paper presents an automatic approach that captures and classifies photo-realistic views at the windows in a 3D photogrammetric city model. First, by triangulating the window geometries from geo-matched 3D photogrammetric and 2D digital maps, the rich window semantics are registered to the 3D models. Then, the similar window views are visualized in batch with an appropriate focal length and field of view. Finally, the view at each window is analyzed and classified through transfer learning automatically. We applied the proposed approach to the 3D model of Hong Kong Island and found satisfactory results for identifying nature scenes or urban scenes. Once massively adopted, the presented approach can offer novel geographic indicators for billions of urban inhabitants and the Architecture, Engineering, Construction, and Operation (AECO) industry.