A framework for reliable traffic surrogate safety assessment based on multi-object tracking data

Multiple object tracking (MOT) systems enable the recording of traffic situations and the movements of road users in high detail. These data form the basis for safety-related analyses such as surrogate safety assessment (SSA), which often involves detecting, quantifying, and analysing conflict situations. Due to the rarity of actual conflicts even occasional data errors can significantly affect SSA outcomes. Consequently, high-quality data are essential. However, a gap remains between MOT and SSA research, particularly regarding the impact of data quality on the reliability of SSA results. This study addresses that gap by proposing a framework that explicitly accounts for the effects of data quality to ensure reliable SSA outcomes. Since it treats the data-generating MOT system as a black box, the framework can also be applied by practitioners using historical datasets or in cases of restricted access to the MOT system. Using the surrogate safety measures (SSMs) time-to-collision (TTC) and post-encroachment time (PET), we illustrate how data inaccuracies affect conflict detection and show how the proposed framework can reveal critical data limitations. We also demonstrate its ability to identify the need for data correction methods and to analyse the effects of such methods on SSA outcomes. Finally, our findings underline the importance of scenario-specific data evaluation for ensuring reliable SSA results and suggest that robustness against data inaccuracies should be considered a key criterion when selecting SSMs.