Automated Vision-Based Detection of Impairment Through Divided Attention Psychophysical Tests

Abstract

Divided attention psychophysical tests are one of the main tests from standardized field sobriety tests (SFSTs) that drug recognition expert (DRE) officers employ to detect impaired drivers and to investigate the type of consumed drugs. Two well-known divided attention psychophysical tests are One Leg Stand (OLS) and Walk and Turn (WAT), which are commonly used by officers to make a decision on the status of the drivers. As this decision might be considered by courts for further investigation, the purpose of this letter is to design an automated impairment system to remove the subjectivity of SFSTs by helping officers make accurate determinations of sobriety and to serve as evidence for proving the correctness of the officers’ decisions in the courts. In this letter, a vision-based system is introduced and implemented to automatically detect impaired subjects using various feature engineering and machine learning algorithms, which were performed on the OLS and WAT videos obtained from 34 volunteer participants. Based on the results, the Random Forest classifier showed the best performance for impairment classification, achieving results comparable to those of DRE officers. Furthermore, the OLS-right features are the most relevant compared to the WAT features for the final classification.