Analysis of urban growth and decline by streetview images using deep learning

Abstract

In urban studies, analyzing urban growth and decline is a good example of demonstrating urban change. Over the past several decades, various approaches to urban change detection have been developed, most notably urban change detection using satellite imagery. However, existing approaches cannot easily detect and analyze specific changes at the street level. Hence, we seek to use street-level panoramic images of Google StreetView to identify changes and to develop a new way to evaluate the growth and decline of cities. After collecting Google Street View images year by year, we undertake an object detection process using the open-source software Tensorflow based on a neural network. Because the detected objects show changes in urban areas, we score the objects or factors detected street by street by evaluating them. Objects or factors detected at the street level show how far the street has advanced, grown, or declined. In fact, streets with numerous growth factors had high scores and frequent changes, and streets with declining factors had low scores of changes. The results of this study are expected to contribute to city policy planning and urban planning based on sociological research results, along with forecasts of future urban changes.