Abstract:
Surrogate safety analysis refers to a large class of methods to evaluate safety at any site without waiting for collisions. A core requirement of any safety analysis method is the calculation of reliable safety indicators, which are mostly computed at a given time with the use of the positions and speeds of the two interacting road users. These positional data can be obtained automatically with video sensors and computer vision techniques. Therefore, automated safety analysis based on video data has gained the interest of safety analysts. However, automatic, video-based methods for surrogate safety analysis generally consider road users as points. Hence, the geometric representation of the road user volume (the outline) can affect the accuracy of the calculated indicators and consequently the safety evaluation. This study investigated geometric representations of road users for the automated computation of safety indicators. In particular, they were compared in the measurement of postencroachment time. This investigation relied on a real-world case study at two sites with two camera configurations, the first with an almost overhead view and the second with a much lower camera angle, which was more common in practice. The results show that representing the road user as a convex hull with extracted features is most accurate, with a root mean square error equal to 0.25 s for computation at the first site with an overhead view. With a lower camera angle, the advantage of that method disappeared and a simpler circular representation with a fixed radius showed the best results.
