Facial Expression Recognition and Spatial Analysis in Urban Environments: An Exploratory Proof-of-Concept Study

Facial Emotion Recognition (FER) has transformed human–computer interaction, yet its application to understanding relationships between physical environments and human emotional responses remains an emerging area within urban studies. This study presents an exploratory proof-of-concept framework that integrates FER-derived facial expression classification with computational spatial analysis to investigate associations between urban spatial attributes and inferred affective responses in a real-world high-density residential setting. Employing a combination of pretrained deep learning models for feature extraction and pseudo-label generation, together with a custom-trained multimodal architecture, the framework was tested through a case study in Tung Chung, Hong Kong, with 15 participants traversing a 12-segment walking route. Two primary contributions are offered in this study. First, it provides preliminary evidence for the potential utility of FER as a measurement approach in real-world physical environments, while transparently documenting the significant limitations of model-inferred emotion labels, including moderate agreement with human annotation (κ = 0.63 for binary valence) and poor performance on specific negative expression categories. Second, the study presents a preliminary predictive model that identifies spatial attributes—particularly light, greenery, and vertical movement—as consistent associates of FER-inferred valence under participant-independent validation (leave-one-participant-out (LOPO) F1 = 0.60), although generalizability across spatial contexts remains limited (leave-one-segment-out (LOSO) F1 = 0.52). All findings require replication with larger samples, validated ground-truth emotion measures, and multiple independent sites before practical inferences can be drawn. This research offers an exploratory methodological contribution to the nascent field of spatial emotion recognition and a foundation for more rigorous future investigations.