Identifying cognitive state from eye metrics

Abstract

Introduction: This paper describes a new approach for identifying cognitive state by using information obtained only from the eye. Data are collected from cameras mounted on a lightweight headband. A set of eye metrics captures essential eye information from the raw data of pupil size and point-of-gaze. The metrics are easily calculated every second, so that the entire set of metrics can be computed in real time.

Methods: Three studies provide empirical evidence to test whether the eye metrics are sufficient to discriminate between two different cognitive states. The first study examines the states of relaxed and engaged in the context of problem solving. The second study looks at the states of focused and distracted attention in the context of driving. The third study inspects the states of alert and fatigued in the context of visual search. Two statistical models are used to classify cognitive state for all three studies: linear discriminant function analysis and non-linear neural network analysis. Data for the models are eye metrics computed at 1-, 4-, and 10-s intervals.

Results: All discriminant function analyses are statistically significant, and classification rates are high. Neural network models have equal or better performance than discriminant function models across all three studies.

Discussion: The seven eye metrics successfully discriminate between the states in all studies. Models from individual participants as well as the aggregate model over all participants are successful in identifying cognitive states based on task condition. Classification rates compare favorably with similar studies.