This paper presents a novel method for estimating bicycle and motorcycle ridership in 185 cities worldwide, leveraging Google Street View (GSV) imagery and deep learning (YOLOv4 model). Fine-tuned on images from six cities, the YOLOv4 model detected bicycles and motorcycles with an average accuracy of 89%. A beta regression model, with bicycle and motorcycle ridership as the dependent variable and the log-transformed count of bicycles and motorcycles detected in GSV images as the explanatory variable, was developed to predict bicycle and motorcycle ridership with an R² of 0.614 and 0.612, respectively. Prediction accuracy was generally high, with exceptions in some cities (e.g., Utrecht and Cali). Using this model, we also provided estimates for 60 cities where up-to-date ridership data was unavailable. This study leverages computer vision alongside existing data sources to provide insights into transportation modes and activities.
