Inspiration
Driving on 6-7 hours of sleep doubles the risk of getting into a crash, and driving on less than 5 hours of sleep doubles it again. After just 20 hours without sleep, a driver’s reaction time is comparable to those with a blood alcohol content (BAC) of 0.08% – well above the legal limit. Additionally, it is estimated that 6,000 deaths every year are attributable to driving under sleep deprivation (source).
Some metrics, which are directly measurable by eye gaze alone, have been shown to have a significant difference between well-rested and sleep-deprived test subjects, including gaze–target synchronization and binocular coordination and increased visual reaction time. We take steps to create an accessible, efficient test to quantify sleep deprivation before driving.
What it does
We have built a more accessible way of measuring cognitive impairment due to sleep in the form of a reaction time test. The user is presented with a red bar on either side of the screen, which switches sides randomly. They must keep their eyes on the bar, and from computer webcam data, we are then able to calculate the reaction time. Slower reaction times are correlated with more severe cognitive impairment.
How we built it
We created an eye tracker with OpenCV and used it to estimate iris location. This provides a JSON file containing the relative iris position for the left and right eyes at a 15 Hz sampling rate along with a log of which visual stimuli were shown to the user and at what time. We analyze this data by taking the 1 second of eye recording data after each stimulus, identifying the point of maximum iris speed, and treating that as the reaction time. Then, we aggregate this across ~8 trials to reduce variance, and generate a PDF report using pdfkit and wkhtmltopdf.
Accomplishments that we're proud of
We created something technically advanced that has potentially interesting use cases.
What we learned
We learned about the correlation of eye-tracking measures with reaction time, attention, and general cognitive impairment, as well as the technical components of recording and processing raw eye movement data.
What's next for Cognivision
During this project, we mainly tested on our moderately sleep deprived team members. In the future, we could collect data across different demographics (age group, gender, etc.) in all states of impairment to better determine which reaction times should be correlated with which states. In addition, since reaction time might vary greatly across individuals, we could calibrate an individual’s reaction time and base future predictions on previous results. We could also perform tests for other cognitive metrics such as memory, binocular coordination, and visual attention, to obtain an even more accurate picture of cognitive impairment.
Log in or sign up for Devpost to join the conversation.