Assessing In-Motion Urban Visual Perception: Analyzing Urban Features, Design Qualities, and People’s Perception

The paper proposes a machine learning and computer vision methodology for evaluating urban visual perception during pedestrian movement. The research combines semantic image segmentation, Place Pulse 2.0 perception datasets, Google Street View imagery, and supervised machine learning models to analyze how urban physical features and spatial design qualities influence people’s perception while walking. The workflow integrates PSPNet semantic segmentation, SVM-based perception prediction, and Pearson correlation analysis to investigate relationships between urban morphology, perceived openness, imageability, enclosure, complexity, and subjective perceptions such as beauty, safety, liveliness, boredom, and depression. The study highlights how directional and panoramic visual fields produce different perceptual outcomes and demonstrates the role of AI-assisted urban analytics in understanding pedestrian experience and informing human-centered urban design.