Urban Comfort Perception Under Induced Emotional Conditions: A Multi-Method Analysis of Architectural and Streetscape Imagery Using Fractal Analysis, Self-Report, and Eye-Tracking

This pilot study examines how experimentally induced emotional states interact with the visual properties of urban environments to shape comfort perception. A controlled laboratory experiment was conducted with 17 participants assigned to one of four emotional conditions (Fear, Anger, Sad, Happy) through audio-visual induction. Participants evaluated 73 building façade and 42 pedestrian streetscape stimuli from three urban areas in Kitakyushu, Japan (Wakamatsu, Tobata, Mojiko) using a multi-method framework combining fractal analysis (D, Λ), six pedestrian visual metrics, webcam-based eye-tracking (Visual Attention Score, VAS), and self-reported comfort votes. Emotion induction was effective for Fear and Anger groups and partial for Sad and Happy groups, with the latter attributable to experimental fatigue. Cross-method correlation analysis revealed that fractal dimension D significantly predicted comfort vote consensus (Spearman r = 0.369, p = 0.013), while VAS showed no significant relationship with comfort votes (r = 0.097, ns) or with fractal dimension (r = 0.015, ns), confirming that visual attention and comfort preference are independent dimensions. For building façades, the ‘Complex but Organized’ fractal profile (D ≥ 1.70, Λ < 0.60) was the consistent comfort driver across all emotion groups. For pedestrian streetscapes, low spatial enclosure and spatially integrated tree canopy were the primary comfort predictors. Multi-method synthesis identified five empirical paradoxes and three design principles: (1) target D ≥ 1.70 with Λ < 0.60; (2) prioritize spatially integrated canopy over visible greenery quantity; and (3) leverage civic legibility as an independent comfort pathway. These findings support the development of emotion-independent frameworks for urban comfort evaluation. Replication with larger, more diverse samples is recommended.