AI and Computer Vision for Park Studies in Abu Dhabi: Understanding Urban Parks Through Large-Scale Image Analysis

Author Information

Amina Menaa, Sorbonne University Abu Dhabi, United Arab Emirates
Lama Tarsissi, Sorbonne University Abu Dhabi, United Arab Emirates
Xavier Fresquet, Sorbonne University Abu Dhabi, United Arab Emirates
Marianne Cohen, Sorbonne Université, France

Abstract

We present an interdisciplinary study that applies computer vision to analyze urban parks in Abu Dhabi. Leveraging a dataset of approximately 12,000 user-generated images sourced from Google Maps, we built a pipeline to clean, process, and interpret visual information from green spaces, encompassing metadata filtering, geolocation curation, duplicate removal, and semantic outlier detection. Using YOLOv8* object detection, spatial density maps, temporal trend analyses, and behavioral heatmaps, we examine how parks are used and perceived by the public. Our findings reveal distinct spatiotemporal rhythms of park use, including strong evening peaks reflecting climate adaptation and weekend surges tied to family-oriented activity. Frequently detected objects such as palm trees, other trees, flowers, and fountains capture the ecological character of these spaces, with natural elements consistently dominating visual prominence and serving as proxies for the social and restorative functions parks fulfill. Spatial density maps further identify certain parks as primary centers of activity. We also reflect on the challenges of applying pretrained AI models to heterogeneous, user-generated data, including sampling bias and model transferability. This work contributes to broader Nature-Based Solutions (NBS) efforts and supports evidence-based urban green space planning in rapidly urbanizing contexts.

Save as a Favourite