I – Introduction

1Smiley & Brookhuis (1987) conclude that some 90 % of all traffic accidents can be attributed to human failure. Therefore, the prévention or réduction of traffic accidents requires behavioural change among the responsible road users. For this to occur, firstly the mechanisms of accident causation have to be understood and the specifie behaviours contributing to accidents have to be identified. Then, countermeasures have to be devised and introduced to prevent those behaviours contributing to accidents, without eliciting undesirable side-effects, e.g. unwanted displacement behaviours. Recently, combined ergonomie and engineering approaches to both hazard assessment and the indication of drivers’ performance limits have developed into research and development (R & D) of new and relevant (primary) safety measures. Brookhuis & Brown (1992) argue that an ergonomie approach to behavioural change via engineering measures, in the form of electronic driving aids, should be adopted in order to improve road safety, transport efficiency and environmental quality.

2The Commission of the European Communities (CEC) proposed a R & D programme in the field of road transport informatics and telecommunications. The objectives of the programme were defined as the improvement of road safety, transport efficiency and environmental quality. In 1988 a research programme, Dedicated Road Infrastructure for Vehicle Safety in Europe (DRIVE), was formally adopted by the CEC for an initial period of three years. DRIVE envisages a complété general European road transport environment in which individual drivers are better informed and monitored ; and in which ‘intelligent’ vehicles communicate and cooperate with each other, the road users and the road infrastructure itself. DRIVE seeks to create favourable conditions for the development of this integrated road transport environment through collaborative R & D in the field of information technology and telecommunications applied to road transport. It seems likely that this comprehensive concept reduces political objections to the introduction of many safety measures which include constraints on individual drivers’ freedom. Any measure that marginally constrains individual road user behaviour in order to improve safety is much more acceptable if it can be shown incidentally also to reduce traffic congestion and air pollution.

3The main purpose of implementing electronic driving aids in vehicles, however, should be to improve road safety. This can be attained, for instance, by detecting decrements in driver performance. It is repeatedly shown that performance decrements in general may be due to common factors such as alcohol, medication, illness, temperature, sleep derivation, fatigue, and with respect to driving performance specifically, driver underload or overload (for an overview see Thomas et al., 1989). Certain projects within the DRIVE II programme aim to detect inadéquate vehicle control under conditions where the driver’s cognitive, perceptual and motor abilities may become impaired (Le., accident risk is increased). The DETER (Detection, Enforcement and Tutoring for Error Réduction) project is an example of this kind.

II – The Deter project

4The aim of the DETER project is to demonstrate the feasibility of using advanced transport telematics (ATT) technology in order to assess driver behaviour. The resuit of this assessment would be the provision of warning, tutoring and other feedback messages and to register serious driver errors or law violations within a limited set of critical situations. A technological distinction has been drawn between those ATT Systems based on-site at the road-side, those based in-vehicle and those integrated Systems which combine information from both on-site and in-vehicle sources. On-site Systems mainly focus on traffic law violations, for example overspeeding and red light violations whereas in-vehicle technology relies on monitoring vehicle parameters. Naturally an integrated System should achieve optimal potential safety benefit by combining both sources of information and resenting the appropriate feedback to the driver.

5In the development phase threc separate Systems are in operation, on-site testing of detection, tutoring and enforcement Systems, their legal-administrative and road user acceptance, behavioural, cost-benefit and safety assessment, and in-vehicle behaviour assessment Systems, based on vehicle sensor information and finally the driver impairment monitoring System. The first stream is located at the ilot site in Norway, the second carried out in several laboratories including simulators and in the field, separately and in conjunction.

6The assessment of the legal-administrative issues and social acceptance of operational on-site warning, feedback and enforcement Systems has started by means of different experimental studies, both on-site and in-vehicle. Target behaviour concerning speed, distance keeping and vehicle load are evaluated in the on-site project in which both subjective measures (that is reported behaviour by questionnaire) and objective measures (that is registered behaviour) are included. The development of a prototype in-vehicle System capable of detecting traffic law violations, providing tutoring and feedback and registering the traffic violations committed by the driver, is put to tests in simulators and real traffic. The development of a prototype of an in-vehicle System capable of detecting deviations from the driver’s normal State of vigilance, registering occurrence of the impairment and providing feedback to the driver is an extension of earlier work in the DRIVE I project V1004 (DREAM). In this project, a feasibility study has been carried out for monitoring driver status (Thomas et al, 989). The information that was evaluated to this end was derived from a literature study and experimental research. It was concluded that the combined measurement of the driver’s physiology and behaviour has demonstrated that it is feasible to develop a monitoring device on the basis of unobtrusive vehicle parameter alone. The driver actions, as measured from vehicle parameters, that seemed most promising were aspects of steering wheel movements, variability in latéral position, reactions to lead vehicles and perhaps speed management.

III – Driver impairment monitor

7This contribution concentrates on the development of the in-vehicle driver impairment monitoring System (DIM), as proposed by Fairclough et al (1992). It concerns the development of a prototype System, capable of detecting driver impairment from vehicle sensors alone. The DIM is divided into three functional units, the vehicle sensors that collect instantaneous driving data, the neural net System that analyses and interprets these data and the storage/retrieval device which is used as a template of normal driving behaviour. The flow of information from vehicle sensor to interrelation proceeds as follows. Data are collected from the vehicle sensors and recessed into the form of critical vehicle parameters, i.e. those driver actions that are indicators of psychological impairment. The data are interpreted by a neural net algorithm with reference to the storage/retrieval design that contains a record of the individual driver’s normal style of driving. The output from the neural net System diagnoses either normal or abnormal driving. In the latter case a separate part of the System will select and produce tutoring message(s), or start enforcement. In the recent DETER project several studies hâve been carried out that contribute to a first approximation of the détermination of critical driver actions and the assessment of their critical values with respect to psychological impairment. Critical driver actions, as derived from DETER and earlier studies, are :

1. lateral position handling,
2. steering wheel handling,
3. speed management,
4. headway control,
5. pedal use.

8Critical values are of two types, absolute and relative. Absolute critical values refer to levels of measured driver actions that would imply a direct, acute danger of accident involvement to the vehicle and driver. For example, driving at less than 0.2 second time-headway to a lead car whereas a minimum human reaction time in laboratory circumstances is around 0.2 second. Relative critical values are individual and refer to decrements in driver actions that indicate psychological impairment, without direct relationship to likelihood of accident involvement. For instance, if a person, who is inclined to avoid risks and normally drives at 2 seconds headway to a lead vehicle, now drives at 1 second, he is probably not in an optimal condition.

1. Lateral position handling

9Since around 1980 lateral position statistics are developed and repeatedly applied in on-the-road tests for measuring effects of drugs on driving performance (O’Hanlon et al, 1982, Brookhuis et al., 1993). In particular the ability of the driver to control weaving of the car, measured as the standard déviation of latéral position (SDL), is a very sensitive indicator of drug-induced driver impairment. However, drugs differ greatly in that potential as is clearly shown in figure 1. The equivalent effect of blood alcohol concentra- lion (BAC), also deicted in this figure is derived from a separate study specially undertaken to establish the effects of three levels of alcohol, relevant in the sense of legal limits in different locations in Europe, i.e. 0.05 % and 0.08 %, and 0.1 %, which used to be the legal limit in several places in the world. From the relationship between BAC and the ability to control weaving, expressed as SDPL, it follows that the critical increase in SDPL should be in the range of 2.5 to 4 cm.

Figure 1. Mean changes (relative to placebo) in SDPL after two consecutive administrations of (mostly benzodiazeine) hypnotics, in the morning and afternoon, referred to separately measured changes with indicated BAC

10It is also worth while to put sonie effort in interpreting SDPL in terms of traffic safety. The probability of lane edge exceedance can be taken as a (surrogate) measure of potential accident involvement. The relationship between probability of lane excellence and SDPL is by Allen & Stein (1987). From this relationship follows that starting at an absolute SDPL of around 25 cm the likelihood of lane exceedance rapidly increases into a substantial proportion, inevitably leading to accidents in due course.

1. Steering wheel handling

11A number of steering wheel parameters hâve bcen developed in the past, indicating decrements in driver performance (e.g. McLean & Hoffmann, 1975), for instance, steering wheel reversais, number of zéro crossings, standard deviation, amplitude range, integrated amplitude over time, jerk (rise time), frequency characteristics, and recently more sophisticated combinations of some of the afore-mentioncd parameters (see Planque et al., this volume).

12Examples of the latter ones are the Y-functions, developed by Renault (Petit et al., 1990), and the S-function, developed by INRETS (see Fairclough et al, 1992), shown in figure 2.

13Careful field experiments including physiological measurements for independent establishment of signs of decreasing vigilance lead to the determination of thresholds in these respective functions with regard to the turnover from vigilant into non-vigilant.

Figure 2. Steering wheel function (S-function) under higher and lower vigilance conditions related to theta activity in the EEG power spectrum

3. Speed management

14Several authors in the past decades mentioned « deliberate » speed variations as ways to suppress upcoming fatigue. Drivers increase their vehicle’s speed with the objective to increase their (feelings of) arousal, only temporarily of course, after which they let their vehicle’s speed drop again. In the series of prescription drug experiments carried out by TRC, significant increases in standard deviation of speed was demonstrated, predominantly in cases of severe sédation. It is also reported that drivers show speed drops whenever their attention is diverted seriously, which is demonstrated, for example, in case of telephoning while driving (see Brookhuis et al., 1991). It seems reasonable (but not irrefutable) to suppose that drivers normally keep their speed steady at a level of their choice, at least while circumstances allow such freedom, which will in many cases be the local speed limit. Exceeding the speed limit by 10 %, not surprisingly the margin the police « allows » in most situations, seems a reasonable criterion with respect to standardisation. The standard deviation of speed that accompanied high SDPL after some prescription drugs and was significantly different from placebo by itself, amounted to a 5 % increase.

4. Headway control

15A certain distance between cars bas to be kept in an absolute sense to be able to react adequately, which is dépendent upon the speed (Fuller, 1984, De Waard et ai, 1991). Therefore, a criterion for time headway, a measure of distance relative to speed, should be feasible to be assessed as absolute threshold for safe driving from the literature, whereas a relative criterion seems still to be determined.

16Recently the standard driving test, developed and in use by TRC, bas been expanded to include a car-following test to study the reaction of the driver to driving behaviour (speed management) of other road users with respect to distance keeping or headway control (Brookhuis et ai, in press).

17Drug sedation or activities such as using a car telephone have been shown to impair these reactions. From several experiments a baseline criterion could be derived for preliminary (absolute) threshold, as well as a relative criterion based on alcohol and prescription drug affected increases pedal use.

18Although edal use is one asect and sometimes the lever mechanism of speed management, in a sense, it is treated searately here. Fairclough et al. (1992) studied brake edal use after alcohol and found significant différences after low amounts of alcohol (BAC < . 05 %), compared to sober trials. Subjects decelerated later and more forcefully as they approached junctions. Also approaching and taking bends is carried out differently after relatively low amounts of alcohol.

IV – Conclusion

19So far, several studics within the DETER project hâve served to demonstrate that analysis of driving behaviour by combining and interpreting vehicle parameter data reveals evidence of driver impairment, either or not validated by concomitant physiological measurement. This has been shown in studies with (low amounts of) alcohol, sleep deprivation and prolonged driving, evidencing the potential of the driver impairment monitor’s sensitivity to distinguish stages of impairment from real insidious détériorations in performance. Some preliminary standards for the determination of impairment are now developed, shown in table 1, both in the absolute sense and as relative criteria to be established for each individual assessed as deviation from his or her normal behaviour. For the latter, the neural net algorithm will have to be trained and tuned to idiosyncratic driving patterns that will be stored on personal smart-card. Whenever the driver enters the car and inserts the smart-card, the driver impairment monitor will then be able to recognise the driver and use the stored information to determine the relative criteria. In this way, the System should also be able to assess driver impairment at an early phase of the ride, for instance in case of alcohol use.