Despite persistent efforts at the local, state,

and federal levels, alcohol-impaired crashes

still contribute to approximately 30% of all

traffic fatalities. Although enforcement and

educational approaches have helped to

reduce alcohol-impaired fatalities, other

approaches will be required to further reduce

alcohol-related fatalities. This paper

describes an approach that detects alcohol

impairment in real time using vehicle-based

sensors to detect alcohol-related changes in

drivers’ behavior.

Data were collected on the National

Advanced Driving Simulator from 108

volunteer drivers. Three age groups (21-34,

38-51, and 55-68 years of age) drove

through representative situations on three

types of roadways (urban, freeway, and

rural) at three levels of blood alcohol

content (0.00%, 0.05%, and 0.10% BAC).

Driver control input, vehicle state, driving

context and driver state data, individually

and in combination, reveal signatures of

alcohol impairment. Algorithms built on

these signatures detect drivers with BAC

levels that are over the legal limit with an

accuracy of approximately 80%, similar to

the Standardized Field Sobriety Test (SFST)

used by law enforcement. Each of the three

algorithms combined information across

time to predict impairment. The time

required to detect impairment ranged from

eight minutes, for complex algorithms (i.e.,

support vector machines and decision trees

applied to relatively demanding driving

situations), to twenty-five minutes for

simple algorithms (i.e., logistic regression).

Timely impairment detection depends

critically on the driving context: variables

specific to the particular driving situation

result in much more timely impairment

detection than generic variables.