Which street side catches the eye? Techniques for analyzing pedestrian gaze in urban landscapes

Abstract

In response to the growing demand for more advanced tools in the eye-tracking field, this thesis enhances the MYFix approach [1] by incorporating advanced data analysis techniques for automated interpretation of eye-tracking data. The thesis explores the development of two innovative approaches aimed at enhancing the automation of eye-tracking data analysis. It introduces the polar angle and quadrant approaches, each designed to systematically determine whether an individual’s gaze is directed towards their side of the street or across to the opposite side. Although this research primarily focuses on urban pedestrian environments, the applications of these methods extend to other fields like urban planning. The polar angle method calculates the angle between the pedestrian’s gaze direction and a reference point, offering a quantitative measure of gaze directionality. The quadrant approach segments the visual field and analyzes the location of gaze point and the distribution of features, focusing on the road in this particular study. The manual annotation and quantitative analysis of 50 frames from an eye-tracking video revealed a precision of 30% for the polar angle approach and 92% for the quadrant approach. The research identifies potential improvements and suggests future avenues for refining these approaches, such as evaluating them in a more diverse set of scenarios, increasing the granularity of the quadrant analysis and incorporating depth of focus to enhance the understanding of foreground versus background engagement. Additionally, the thesis conducts a comprehensive descriptive analysis of eye tracking data, which includes metrics such as label frequency, mean fixation duration, dwell time, transition matrix, and temporal visualizations, revealing the complex dynamics of gaze behavior in urban settings. This groundwork could be further advanced by developing a software toolkit that encapsulates all steps of the data analysis pipeline, thereby standardizing and streamlining the process to make advanced gaze analysis techniques more accessible to researchers and practitioners in different fields.