AI has the potential to guide urban planning decisions.
Accurate measurement of the quality of urban spaces is a critical aspect in creating effective policies that tackle the various challenges faced by urban dwellers such as poverty, infrastructure, transportation, health, and safety. However, traditional methods of collecting socioeconomic data like crime rates, income levels, and housing conditions through occasional citizen surveys are inadequate, as they are infrequent, expensive, and rely on human perception, leading to an outdated picture of the conditions at a neighborhood level.
The use of AI in this field is rapidly gaining popularity, with researchers experimenting with satellite imagery to quantify urban sprawl and machine learning projects to generate large-scale mappings of poverty, wealth, and income in developing countries. Despite these advancements, the physical landscape within urban environments and how it changes over time has yet to be fully explored.
According to Andrea Vallebueno, "there is no adequate measure that documents the quality of the urban space, its change over time, and the spatial inequality it presents." Andrea worked with her co-author, Yong Suk Lee, to fill this gap by using high-frequency Google Street View images and constructing panel data at the street segment level, making them the pioneers in this field.
Andrea Vallebueno is a Research Fellow of Computational Science at Stanford University’s Regulation, Evaluation, and Governance Lab.
Andrea’s passion for using data science for social-good applications led her to explore the use of machine learning and vision AI. With an economics background and a Master’s Degree in Data Science from Stanford, Andrea has been using YOLOv5 for about a year and a half.
As a child in Mexico City, Andrea was acutely aware of the dramatic disparities between neighborhoods like Santa Fe, where a large concrete wall separated the rich from the poor. As a data scientist researcher, Vallebueno became concerned about how traditional economic data overlooks these extreme differences, obscuring the indicators of inequality and urban decline. She realized that with the growing influx of people moving into cities worldwide, the lack of detailed data would only become a more pressing issue.
Andrea and Yong aimed to create an objective measure of urban decay. They used object detection in Google Street View imagery to capture eight urban features that are indicative of urban decay. They chose YOLOv5 due to its inference speed and the use of contextual information, which was crucial for their use case.
The trained model was used to run inference on 114,000 street view images from different neighborhoods in San Francisco, Mexico City, and South Bend. The detections of the eight attributes were aggregated at the street segment level to generate indices of urban decay and measure the change in the incidence of urban decay over time.
Andrea and her team found YOLOv5 to be incredibly easy to work with, with the majority of their time spent on curating their dataset and training their models. They appreciated the integration with experiment tracking tools, and the automatic learning of bounding boxes, which made the process much more accessible.
Andrea and her team are excited to expand their measure of urban quality to include positive attributes of the physical urban environment and test the performance of these indices in a diverse set of urban neighborhoods.
Visualization of the set of model detections of tents/tarps used as homeless dwellings over time in the neighborhood of Tenderloin, San Francisco.
For those new to AI, Andrea recommends finding a problem or research question that they are passionate about and going through the full AI life cycle. She believes this is one of the best ways to build intuition and understand the limitations of their model.
AI is increasingly becoming a crucial tool for universities and researchers, as it enables them to explore and understand complex data sets, making their findings more accurate and reliable. By leveraging AI, researchers can create a more comprehensive understanding of urban spaces and the challenges urban dwellers face, leading to better policies and solutions.
Thanks for reading about Andrea’s experience with YOLOv5! Would you like to share your experience? Tag us everywhere @Ultralytics with your very own Ultralytics YOLO use case and we’ll promote your work to the ML community.
Begin your journey with the future of machine learning