Analysis and Simulation Study of Stochastic Time-To-Collision as a Severity Measure in Traffic Security

Sammanfattning: Traffic accidents are extremely rare, creating the need for surrogate methods for safety analysis that makes efficient use of the information provided by traffic conflicts, which also are limited in availability. The severity measure time-to-collision (TTC) in combination with extreme value theory have so far been one of the primary measures used to infer traffic safety levels, but it relies on unrealistic assumptions that results in severity measures that do not always agree well with observed danger. Stochastic TTC has been proposed as an alternative, which replaces the constant velocity trajectories used to define collision course with naturalistic ones, resulting in a distribution of potential TTC values. The main focus of this thesis is to find a way to mathematically model such data. Stochastic TTC is conceptualized within the framework of mixed distributions, and equations allowing for extreme value theory to be used in such a context are derived. A second point of focus is on presenting a new method for estimating the collision probability, allowing for separate estimation of collisions that occurred with and without attempts at evasive action. Also, the effects of data-transforms were investigated in a simulation setting, which proved highly useful in reducing the tendency for underestimation which seems to be a common problem when extreme value theory is applied to traffic data, and overall improving accuracy. Stochastic TTC and the proposed methods were also tested in this simulated traffic environment, which showed that stochastic TTC can work at least as well as regular TTC, but in order for more significant differences to be seen, the simulation would have to contain more variable road user behavior and more curved trajectories.