This content is not included in your SAE MOBILUS subscription, or you are not logged in.
Combination of a Statistical and a Case-By-Case Approach for Accident Situations Classification
Published May 23, 2004 by Society of Automotive Engineers of Korea in South Korea
Personal injury road traffic accidents are often classified according to a single criterion. For example, accidents can be distributed according to collision type (head-on, rear-end, side, rollover), road geometry (crossing collisions at junctions, accidents on straight roads, loss of control in bends), vehicle configurations (single vehicle accidents, car-to-car collisions, collisions with obstacles, etc.) or even driving situations (overtaking, change of direction, loss of control, U-turn, left-turn, right-turn, parking accidents, etc.). In addition, French researchers have also proposed an accident classification according to the so-called scenario concept taking into account the complexity, the diversity and the chronology of accident events. Accident classifications are currently determined by the objectives of the classification and both the type and volume of information available in the databases.
The paper proposes a 3-step method to accidental situation classification based on both a statistical and a case-by-case approach in order to identify, validate or evaluate safety measures. An accidental situation is defined as a situation with which a driver or a pedestrian is confronted. Consequently, there are as many different accident situations per accident as there are road users involved.
First, a sample of approximately 750 car accidents is extracted from the French LAB Database, built up from the in-depth accident investigations carried out by CEESAR in two small areas in France (about 2,000 km2) from 1995 to 2001. The first analysis is a statistical factorial correspondence analysis and an ascendant hierarchical classification used to separate the accidents into 24 different clusters. The classification involves 27 descriptive or analytic parameters (general circumstances of the accident, contributory factors, pre-accidental situation, road geometry, driver failures, dynamic parameters (speed, acceleration,), drivers' actions, etc.). Then, the accident mechanisms of each statistical cluster are validated and illustrated by the in-depth examination of 5 or 6 accident cases per class.
Finally, a random sample of 500 police fatal accident reports are classified by accident experts to give a representative distribution of fatal accidents which occurred in France in 2002. The arguments for such a method and the main classes are presented and the potential value of the identification and evaluation of safety measures is discussed in the paper.