Safety performance measures: a comparison between microsimulation and observational data

Abstract

Safety performance measures can be obtained either through simulation (based on well specified or calibrated traffic models) or experimentally through observational vehicle tracking data. Accurate calibration of traffic models ensures that simulated measures of safety performance are reflective of “real world” traffic conditions. The microscopic model, for a case study, allows the estimation of road safety performance through a series of indicators, representing interactions in real time between different pairs of vehicles belonging to the traffic stream. When these indicators reach a certain critical value, a possible accident scenario is identified. For the same case study, safety performance indicators are obtained through a video image processing algorithm for vehicle detection and tracking. The accuracy of the algorithm is evaluated with respect to GPS tracking measurements. The algorithm adopts a background subtraction-based approach for vehicle detection in 0.1 second increments. Since this approach is sensitive to background changes (or noise), a median filter technique has been introduced. Individual vehicles are detected and tracked using a region-based approach, whereby a connected zone (or blob) is assigned to each image, which is then tracked over time. In case of overlapping, where the designated blob may correspond to several vehicles, a real time sub-routine is accessed that manually discriminates each constituent vehicle's specific position within the blob. Output from the algorithm application is expressed in terms of several trajectory descriptors over time, such as position and speed. The focus of this paper is on the analysis of road safety from two different perspectives: microsimulation and observational data. In this way it is possible to determine how microsimulation reflects “real” driver behavior and traffic conditions for a given case study.