Pixels, Plants, and People: Affective Evaluation of Urban Green Spaces

Paper Title

Pixels, Plants, and People: Affective Evaluation of Urban Green Spaces

Publication Info

• Topic area: Affective evaluation of urban green spaces using multimodal physiological and visual data.
• Keywords: Urban green spaces, affective evaluation, eye tracking, EEG, vegetation composition, urban design, greenery threshold, multimodal sensing, user perception, urban informatics.

Background and Problem

• Problem / challenge: Urban planners lack scalable, composition-aware tools to predict how urban green spaces will be perceived by users, relying instead on coarse metrics like canopy cover.
• Significance: Understanding the affective impact of urban green spaces is critical for improving well-being, safety, and satisfaction in urban environments.
• Motivation and related work: Prior research has shown benefits of greenery for mood and stress but often relies on single modalities, lacks pixel-level semantic analysis, and does not release open datasets. This study addresses these gaps by linking pixel-level scene composition to multimodal physiological and affective responses.

Solution

• Proposed approach: A lab study combining eye tracking, EEG, and user ratings to evaluate affective responses to urban green spaces, using pixel-level semantic segmentation of urban scenes.
• Novelty:
  1. Pixel-level analysis of urban scenes to link visual composition with affective responses.
  2. Multimodal integration of eye tracking, pupil dynamics, and EEG to assess user experience.
  3. Identification of a greenery threshold (~15%) and specific vegetation elements that enhance perceived pleasantness.
  4. Public release of a dataset with segmented images, physiological responses, and analysis scripts.
• Procedure and key techniques:
  1. Participants viewed 30 urban site images while eye movements and EEG were recorded.
  2. Images were semantically segmented into 14 urban classes and analyzed for vegetation and non-vegetation proportions.
  3. User ratings were collected for valence, safety, satisfaction, and calmness.
  4. Multimodal analysis linked scene composition to gaze patterns, pupil responses, EEG activity, and user ratings.

Results

• Concrete findings:
  • Scenes with ~15% greenery were consistently rated as pleasant.
  • Trees and herbaceous layers contributed most to positive perceptions, while hardscape and vehicular elements negatively impacted ratings.
  • Pleasant scenes elicited smaller pupil dilations, more sustained gaze, and EEG patterns indicative of relaxed engagement.
• Advantage over baselines: Moves beyond generic "greener is better" guidelines by identifying specific vegetation types and proportions that drive affective responses.
• Experiments / evaluation:
  • 27 participants viewed 30 images of urban sites in a lab setting.
  • Eye tracking (150 Hz) and EEG (8 channels, 250 Hz) were used alongside user ratings.
  • Semantic segmentation and greenery analysis were performed on images.
  • Results were validated with an online study of 27 participants from a different cultural context.
• Limitations and future work:
  • Small sample size and static lab-based images limit generalizability.
  • Cultural and environmental factors may influence results.
  • Future work should include real-world or XR validation, mobile sensing, and larger, more diverse samples.

Summary

This study demonstrates that the affective qualities of urban green spaces can be anticipated by linking pixel-level scene composition with multimodal physiological and behavioral data. Scenes with ~15% greenery, particularly trees and herbaceous layers, were consistently rated as pleasant. Eye tracking and EEG data revealed patterns of relaxed engagement and reduced arousal for greener scenes. The findings provide actionable guidelines for urban design and planning, such as prioritizing specific vegetation types and reducing hardscape exposure. The public release of the dataset and analysis scripts supports replication and further research, paving the way for composition-aware, user-centered urban design tools.