Child-friendly city: understanding movement behaviors of children and street environments through children’s mobility data using computer vision

Abstract

This dissertation adds knowledge to understand children’s movement behaviors on streets and their relationships with street environments using computer vision toward a better understanding of creating streets for children. In the concept of child-friendly cities, streets have the potential as places where children can engage in outdoor activities, foster social connections, and derive health benefits. However, rapid urbanization has transformed streets into spaces focusing on vehicles and marginalizing their social functions for children. Inspired by urban planners, cities have invested enormous resources in promoting children’s activities by redesigning urban streets. To improve the quality of streets for children, it is important to investigate children’s use of streets and the characteristics of streets. Previous studies have used varying tools to identify children’s movements and behaviors in urban spaces as well as streets such as interviews, surveys, travel diaries, and systematic observations. With the development of information and communication technology, urban researchers have collected individual movements and behaviors in microscale urban spaces using GPS tracking, mobile data, social media data, and so on. Particularly, recent studies have employed computer vision techniques to analyze human movement behaviors in urban spaces. With the development of vision sensors such as closed-circuit televisions (CCTVs) and machine learning techniques, researchers can support the scalability of traditional studies to analyze children’s movements and behaviors simultaneously and consistently. This dissertation aims to the following key research question: “How do children move on streets and how are their movement behaviors related to street environments?”. Toward that aim, we utilize computer vision techniques to analyze children’s movement behaviors on streets and explore relationships with the physical environments of streets. To the best of our knowledge, this dissertation is the first study to utilize children’s mobility data using vision-based technologies for identifying children’s movement behaviors on streets and analyzing street environments through their movement behaviors. The following research questions are linked to answer the key question, from understanding children’s movement behaviors on streets and their relationships with street environments: 1) How can we collect reasonable children’s trajectories for movement behavior analysis? 2) How do children move on urban streets? 3) Can children’s movement behaviors tell us about the physical environments of streets for children? For the first question, we propose a novel approach to efficiently collect and correct human trajectory data with minimized practical errors in multi-object conditions using vision-based techniques. The results demonstrate the system's ability to significantly reduce practical errors and automatically generate more precise human trajectories compared to current models, enhancing the accuracy of human movement behavior analysis. To answer the second question, we analyze children’s movement behaviors on streets through vision-based techniques by extracting children’s trajectories, calculating movement behavioral features, and classifying the movement behaviors of children. The results demonstrate children’s movement behaviors (walking, staying, and running) on the streets and identify differences in children’s movement behaviors under varying conditions such as day and time by multidimensional analysis. In addition, our automated movement behavior analysis identifies the potential of street environments to positively influence children’s use of streets from the daily street life of children. For the last question, we perform a clustering analysis of children’s movement behaviors on streets and compare them to the physical environments of the streets. In addition, we conduct a statistical analysis of relationships between children’s movement behaviors and street environments. By defining children’s movement behaviors in the context of child-friendly streets, our clustering results identify the street environments that encourage the freedom and independence of children’s movement behaviors on streets such as visual exposure, school zones, and road networks. Our statistical analysis results reveal the influences of the physical environments of streets on different types of children’s movement behaviors such as walking, staying, and running. In addition, we suggest design guidelines for fostering child-friendly streets, established on the results derived from this dissertation. This dissertation shows the feasibility of using computer vision techniques to analyze and understand children’s movement behaviors on streets, validating traditional findings and adding new knowledge about children’s use of streets. In addition, this dissertation adds knowledge to extend the understanding of place-making for child-friendly streets to create better street environments for children’s free and independent movements and behaviors on the streets.