Measuring human alertness from histogram of response time to audio-visual stimuli

Response time test is one of the widely used metrics to measure/validate human alertness. Response time metrics like Mean reaction time (RT) and Mean of 1/RT suffer from lower range and resolution in case of shorter versions of Response time tasks (<10 min). However, small be the fixed test duration, knowledge of it affects the test performance as the subjects might ration their effort over the test period. Therefore, in this article, we fix the number of stimuli (unknown to the participants) instead of test duration and propose a new reaction time metric - called Peak-to- Spread Ratio ( $$ ) - obtained from histogram of response time to audio-video-stimuli (AVS). Experimental results on data collected from 65 subjects confirms that the proposed $$ has better resolution and sensitivity than Mean RT and Mean of 1/RT. We also propose an alertness level index based on a simple quadratic model trained with a set of $$ values to predict the awake hours of a test subject. Experimental results on 25 subjects emphasize that the proposed alertness level index exhibit higher effect-size and correlation with human alertness level over other related states of the art.