Assessment of Drowsiness Using the Johns Drowsiness Scale (JDS): An Objective, Continuous Index Using Eyelid Movement Measures

Abstract

A major issue within the automotive industry has been the development of suitable and valid measures and metrics for assessing driver impairment due to drowsiness (or fatigue) under real-world environments and operating conditions. This study evaluated assessments of the Johns Drowsiness Scale (JDS)—an algorithmic index using objective eyelid movement measures to continuously scale level of impairment resulting from drowsiness. The goal of this project was to relate JDS to other known indices of drowsiness (PERCLOS [PERcentage of eye CLOSure], Observer Rating of Drowsiness [ORD], Karolinska Sleepiness Scale [KSS], standard deviation of lane position, lane deviations, etc.) with the expectation of developing evidence-based measures, criteria, and thresholds to identify the onset and progression of drowsiness. This work also sought to assess JDS’s predictive power in order to characterize if the system provides advance notice of impairment beyond what is currently available. Data for this project was leveraged from a previously collected on-road instrumented vehicle study (unpublished), involving 61 drivers, and following an approach designed to induce drowsy driving episodes. Researchers at the Virginia Tech Transportation Institute sampled data from 25 drivers who completed that study, and their data was analyzed to derive JDS scores. The research did not target sleep-deprived individuals but rather normally rested individuals under the assumption that low-workload environments could be used to induce drowsiness. Driver workload, or task demand, was lowered by asking participants to refrain from engaging in secondary tasks (e.g., listening to the radio, interacting with their cell phone, conversing with passengers, etc.) during the trip; drivers also engaged the cruise control and were asked to minimize lane changes. The route was also specifically selected to be boring with limited traffic. The study protocol was extremely successful at inducing drowsiness under real-world driving conditions, with 82% of drivers indicating they experienced high levels of drowsiness during the trip, achieving a KSS rating of either 8 (sleepy, some effort to keep alert) or 9 (very sleepy). Early signs of drowsiness (driver KSS rating of 6) tended to occur, on average, within the first 30 minutes of the trip and fully drowsy levels (driver KSS rating of 9) within 45 minutes. Results showed that JDS scores were moderately related to PERCLOS, driver and researcher KSS ratings, as well as ORD scores. Time-to-drowsy (TTD) calculations were explored using a variety of measures, including JDS, and triggering thresholds to better understand drowsiness timelines. TTD was found to vary substantially depending on the specific measure of interest and the designated threshold value, with onset averaging as short as 13 minutes to as long as 57 minutes. JDS, PERCLOS, and driver KSS ratings tended to reach threshold drowsy levels in advance of observable driver impairment indexed by driving performance (first drowsy-related lane deviation). For instance, approximately 80% of drivers reached a JDS score of 7 in advance of their first lane departure—on average, 11.66 minutes prior to the lane departure. Similar results were found for PERCLOS, and to a lesser extent driver KSS. Both JDS scores and PERCLOS were found to have relatively high “hit” rates (correct assessments of drowsiness) when using ORD scores as ground truth. JDS yielded somewhat lower miss rates relative to PERCLOS. Study results suggest that JDS is a promising measure that may serve as a sensitive and leading indicator of drowsiness.