The automatic recognition of risks in traffic scenes is a core technology of Advanced Driver Assistance Systems (ADASs). Most of the existing work on traffic risk recognition has been conducted in the context of motion prediction of vehicles. Thus, existing systems rely on directly observed information (e.g., velocity), whereas the exploitation of implicit information inferable from observed information (e.g., the intention of pedestrians) has rarely been explored. Our previous approach used abductive reasoning to infer implicit information from observation and jointly identify the most-likely risks in traffic scenes. However, abductive frameworks do not exploit quantitative information explicitly, which leads to a lack of grounds for physical quantities. This paper proposes a novel risk recognition model combining first-order logical abduction-based symbolic reasoning with a simulation based on physical quantities. We build a novel benchmark dataset of real-life traffic scenes that are potentially risky. Our evaluation demonstrates the potential of our approach. The developed dataset has been made publicly available for research purposes.