A Methodological Approach for Inferring Urban Indicators Through Computer Vision

The physical environment of a community has been proven to have effects on the mental and physical state of a population. As such, the extraction of Urban Indicators (UI) that evaluate the effects of urban development is essential to assert relationships between the surrounding environment and the well-being of a society. Such a relationship, for example, would be the role of green areas in a city on the prevalence of obesity in its population. In addition, these indicators can contribute to the identification and preventive action in risk situations. For instance, a very degraded area with too much waste accumulated may pose serious risks to public health. However, the traditional methods for UI extraction, particularly in the case of physical indicators, are limited due to the lack of standardized data organization and the subjectivity of self-reported responses, while generally being highly resource-intensive and costly. This work aims to create a methodological approach that is capable of applying Computer Vision to automate the extraction of UI, overcoming the limitations of the traditional approaches. This approach takes advantage of tools that offer remote visualization of locations at low cost. Its success depends on the accurate identification of physical urban indicators that can be extracted from an image, and on choosing appropriate Computer Vision techniques to provide the most precise results for such an analysis.