ABSTRACT
Driving simulator has already been applied in detecting the potential risk in road design phase or exploring the deficiency of the road management strategy. Compared to the traditional road safety audit techniques, the driving simulator-based road safety audit technique could provide quantitative data and visual road environment, which will be helpful to outputting accurate conclusions. However, this technique still uses the traditional evaluation indicators and standards, such as speed, acceleration, lane offset, speed difference, velocity gradient, etc. In other words, the new technique just uses driving simulation test to replace the operation speed prediction, but not fully utilized the driving simulator. For instance, the various driving simulator data can support deep analysis. This paper developed a road safety audit technique that was based on the hypothesis of driver behavior consistency. It is assumed that the abnormal driver behavior will appear when the road section has potential risk. Therefore, the risk road section could be found by abnormal driver behavior detection. The abnormal driver behavior is defined as the behavior which is different from the driver's usual behavior. This technique contains two core parts: 1) driver's usual behavior acquisition and 2) abnormal driver behavior detection criterion. The technique can collect each driver's abnormal behavior, so it can reflect driver's individual differences. When each driver's abnormal behavior was integrated in the same road section, the individual driver behavior could be filtered and some common abnormal driver behavior would be left. At last, the new road safety audit technique and the traditional technique were applied in a real project to validate the effect of the new technique. The results showed (1) the new road safety audit technique could detect the risk road sections which the traditional cannot find, (2) the conclusion by the new technique would be much more quantitative than the traditional one, and (3) the visual road environment and the real operation speed in the visual road environment made the conclusion more reliable and reasonable.
