The personal inspection "in situ" of critical points does not permit conclusions to be obtained about the real evolution of the vehicles, since it supposes an odd road element to the normal traffic flow that affects drivers' behavior. Given such diverse problems, it is not surprising that the relationship between accidents and design features is still not known precisely. In addition, it is often necessary for countries to adopt geometric design standards developed elsewhere. The question often arises as to whether or not these standards are able to adequately predict the behavior of local drivers. The "a posteriori" analysis of accident occurrence presents important limitations, since it is impossible to know exactly the real circumstances, regarding vehicle evolution, in which the accident occurred. Besides the number of accidents is reduced and not representative to obtain general valid conclusions. The personal inspection "in situ" of critical points does not permit conclusions to be obtained about the real evolution of the vehicles, since it supposes an odd road element that affects drivers' behavior. The current state of the art of the artificial vision permits it to automatically track the spatial and temporal evolution of vehicles. This makes it possible to model and obtain the vehicle behavior at critical situations: curves, acceleration and deceleration lanes, intersections, weaving links, etc. Safety margins can be determined by comparing the experimentally obtained vehicle evolution model with the one expected from the current design standards. This paper aims firstly to describe an automatic tool for modeling vehicle evolution at critical locations based on artificial vision techniques, and secondly, to obtain a methodology for evaluating and improving these critical points.
Abstract