Abstract
This paper proposes a quantitative model, for description of driver behavior. Three results are introduced for perceptions of unsafe relations between the first and the second party vehicles at intersections; result y stands for existence of the relation, n for non-existence, and u for lack of information. Accident occurrences are expressed in terms of these results and absence of avoidance actions. An acceptable risk region is introduced and the perception results are shown to locate on boundary points along a hyperbola. Superficial human error probability is inversely proportional to a priori probability of unsafe relation. Perception and judgment error ratios derived from the model are consistent with actual data. The maximum risk level turns out to be a very small number around the order of ten to the minus seven. Crossing collision safety devices are required to demonstrate an extremely high alarm reliability for its effectiveness of decreasing the collisions.
