Multi-perspective assessment of greenery visibility from streets, windows, and drones using deep learning and CIM

Urban greenery is 3D volumetric and provides visual benefits to billions of urban dwellers. Although high-rise buildings and drones are expanding urban activities vertically, most greenery studies are limited to street-level observations, relying on 2D/2.5D view imagery. This paper presents a greenery volume visibility (GVV) assessment method by integrating 3D deep learning and city information models (CIMs). A 3D GVV index (GVVI) is defined for three groups of observing perspectives: the street, window, and drone. Semantic segmentation using deep learning detects greenery volumes in photorealistic 3D models, then GVVIs are calculated across all perspective groups. Hot spot maps and disparity analysis comparing the three sub-indices reveal the urban greenery volumes overlooked in traditional ground-level studies. Experimental results on a 4.92 km2 area of Kowloon, Hong Kong, showcased the proposed method’s efficiency (23,490 GVVIs in 256.45 hours); Comparative results between street and drone/window perspectives revealed 37.6% and 31.7% greenery entirely overlooked in traditional ground-level spatial analysis. This paper contributes in three aspects: a 3D GVV index definition, an efficient deep-learning-based geospatial analysis method, and novel quantitative 3D evidence for informed urban greenery planning and maintenance, optimal building designs, and low-altitude drone route planning.