Analyzing usage patterns from video data through deep learning: The case of an urban park

Rapid urbanization, urban density, and COVID-19 effects have highlighted the need for high-quality urban parks within walking distance. A high-quality urban park maximizes a neighborhood's spatial, safety, and social potential, which are key factors to the well-being of its residents. Most studies evaluating urban parks rely on questionnaires, observations, interviews, and post-occupancy methods. These traditional methods are limited regarding the spatial and temporal dimensions as well as the size of the sample under investigation. In this paper, we demonstrate a new approach to evaluating urban parks by focusing on individuals' activity patterns, using big data extracted from city cameras by utilizing deep learning and computer vision. Our case study is a small urban park, Katznelson Garden, located in Or Yehuda, Israel. The imagery data is analyzed in relation to the gender of the parks' users, along with spatial and temporal analysis. Thus, activities during different hours of the day, days of the week, and in various parts of the urban park are identified. The results of our study revealed that females' and males' activity patterns are different and depend on the hour of the day and the type of park characteristics. Moreover, we found that activity levels and patterns varied according to the day of the week. As many cities seek to design better urban parks tailored to their residents' needs, these study findings can contribute to planning decisions by paving the way to customizing the design of urban parks in accordance with the revealed behavior. • City cameras are used to extract big data for urban planning purposes. • Computer vision and spatial analysis provide a new way of evaluating urban parks. • Video data provides rich information about park users' spatial and temporal behavior. • Urban park users' activity patterns vary with the time of day and day of the week. • Females and males use park's characteristics differently.