Loop detector data could bring significant benefits for intersection safety since loops have been widely implemented and the data can be automatically collected in real time. However, the conventional aggregate data ignores individual vehicle information. Such data is difficult to be used to deeply analyze drivers’ behaviors at signalized intersections. By contrast, high-resolution traffic data (event-based or sec-by-sec) provides detailed vehicle arrivals and departures from loop detectors. This data, combined with signal phase changes, could be used to derive vehicle trajectories, which can serve as the foundation for traffic conflict analysis. More importantly, since loop detector data can be easily and automatically obtained in real time with low cost, this could significantly contribute to the implementation of dynamic systems that could inform or alert drivers of emerging and impending hazardous situations.In detail, the intersection safety performance could be estimated based on the following two major functions:

   1) Estimate potential traffic conflicts based on real traffic conditions. This function will focus on estimating both rear-end and crossing (i.e. right-angle) traffic conflicts. The rear-end conflicts include the same-direction conflicts from all approaches, and the crossing traffic conflicts mainly are the conflicts between through movement and conflicting left-turns. To estimate the potential conflicts, first we need to estimate vehicles’ trajectories using high-resolution data. With the trajectory data, we then apply the traffic conflict technique to estimate both types of traffic conflicts at intersections. The overall intersection safety can be evaluated based on a combination of the probability of two types of potential conflicts.

   2) Predict red-light violations. The second function of the proposed system is to identify possible red-light violations, which is a major factor that leads to traffic conflicts. To fulfill this function, we first use high-resolution data collected from advanced detectors, which are located several hundred feet behind stop-line, to predict drivers’ decision of STOP-or-RUN (SoR) at the onset of amber phase. A prediction model developed in our previous research [30] will be applied to estimate drivers’ SoR decisions. After we predict drivers’ SoR decision, we can estimate red-light violations by combining signal phasing information. For example, if a vehicle is predicted as “RUN” and the remaining amber time is not enough for the vehicle to cross the intersection, this vehicle will end up “red-light violation”. The number of red-light-running violations provides some indication of intersection safety. More importantly, this model can be applied to predict red-light violation in real time if using real-time information collected from advanced detectors. The real-time red-light violation information, combined with appropriate control strategies like all-red extension, could be used to reduce potential severe crashes caused by red-light running.