Urban perception by using eye movement data on street view images

Abstract

AbstractUnderstanding the spatial distribution patterns of urban perception and analyzing the correlation between human emotional perception and street composition elements are important for accurately understanding how people interact with the urban environment, urban planning, and urban management. Previous studies on urban perception using street view data have not fully considered the actual level of attention to different visual elements when browsing street view images. In this article, we use eye tracking technology to collect eye movement data and subjective perception evaluation data when people browse street view images, and analyze the correlation between the time to first fixation, duration of first fixation, and fixation frequency of different visual elements and the six perceptual outcomes of wealthy, safe, lively, beautiful, boring, and depressing. Furthermore, this article integrates eye movement data with street view semantic data and introduces a novel method for predicting urban perception using a machine learning algorithm. The proposed method outperforms a comparative model that solely relies on semantic data, exhibiting higher accuracy in perception prediction. Additionally, the study presents a perceptual mapping of the prediction results, providing a visual representation of the predicted urban perception outcomes. As vision is the primary perceptual channel, this study achieves a more objective and scientifically reliable urban perception, which is of reference value for the study of physical and mental health due to the urban physical environment.