Urban transportation infrastructure is a vital component to the societal sustainability and economic growth. Its functionality and safety are key factors for quantifying its performance. It is reported that the cost of road fatalities and injuries are disproportionately high in low- and middle-income countries. Therefore, it is important for these countries to systematically assess road safety and screen road risks to mitigate potential losses and to support sustainable development of the society. However, traditionally methods for such large-scale assessment or screening require a tremendous amount of road and traffic data. To this end, we conducted a pilot project to develop a framework that only takes street view image as the input and can predict the risk level of a road segment. The availability of street view images makes it possible to screen road risks at large-scale. The framework consists of a module to extract road information from images and a module to predict the risk level using the road information. Both modules are machine learning based. The training of the machine learning models is based on multiple sources of data, which is made available through the Development Data Partnership program. We tested the framework in two cities. The accuracy of the final prediction is 72.5%, which shows promising potential.