Computer vision framework for site-scale multidimensional vitality assessment: Lakeside waterfront spaces as a testing ground

Abstract

Spatial vitality is crucial for fostering healthy, socially interactive, economically vibrant, and sustainable urban areas. However, at the site scale, its quantification remains limited by oversimplified metrics and low-resolution techniques, which cannot capture its multidimensional nature or the fine-grained spatial features that drive it. To address this gap, we proposed a computer vision framework (3M-WVA model) that integrates drone video data to dynamically quantify vitality in three dimensions: aggregation, stability, and complexity through object detection, trajectory analysis, and feature extraction. This approach was rigorously tested in twelve lakeside waterfront spaces across a Nanjing urban park, serving as a methodological testing ground for site-scale validation and investigated how built-environment features drive each vitality dimension with spatial specificity. Results demonstrated that facility richness (RF) universally enhanced all dimensions, while area (AREA) showed negligible impacts. Spatial preferences diverge across dimensions: proximity to waterbody (D_water) strengthened aggregation-based vitality, while sheltered spaces (low SVF) with wide water interfaces(W/D) significantly extend stability-based vitality. Notably, green visual rate (GVR) exhibited diminishing returns—beyond a critical threshold (42.5 %), additional greenery showed negligible vitality benefits. This research established an automated framework for site-scale vitality assessment while revealing transferable insights for lakeside design: prioritizing facility density near waterbody and optimizing the layout or form of space over extensive greening. These findings provide planners with evidence-based levers for precision design, thereby advancing sustainable public open spaces development.