Unsupervised Deep Learning to Explore Streetscape Factors Associated with Urban Cyclist Safety

Cycling is associated with health, environmental and societal benefits. Urban infrastructure design catering to cyclists’ safety can potentially reduce cyclist crashes and therefore, injury and/or mortality. This research uses publicly available big data such as maps and satellite images to capture information of the environment of cyclist crashes. Deep learning methods, such as generative adversarial networks (GANs), learn from these datasets and explore factors associated with cyclist crashes. This assumes existing environmental patterns for roads at locations with and without cyclist crashes, and suggests a deep learning method is able to learn the hidden features from map and satellite images and model the road environments using GANs. Experiments validated the method by identifying factors associated with cyclist crashes that show agreement with existing literature. Additionally, it revealed the potential of this method to identify implicit factors that have not been previously identified in the existing literature. These results provide visual indications about what streetscapes are safer for cyclist and suggestions on how city streetscapes should be planned or reconstructed to improve it.