Building a Smartphone Application to Test Alertness before Drivers Take the Wheel

Overview

Objectives/Goals According to the National Highway Traffic Safety Administration (2013), more than 10,000 people die each year in alcohol-impaired and drowsy driving crashes. To prevent such fatal accidents, my goal is to create a smartphone application to check mental agility and motor skills BEFORE a person takes the wheel. The app identifies key biometrics that measure the sobriety/drowsiness of drivers and maps them to smartphone features for instant and accurate detection of alertness. Methods/Materials The app analyzes 3 biometrics: vocal, cognitive, and balance abilities. It then compares the results to pre-recorded baseline data and suggests whether a person should drive. To test for balance, I devised an algorithm to check the balance of a person walking 9 steps heel to toe. It uses the OrientationSensor on the Android phone to collect motion data along the 3 coordinate axes. I tested this 1) when the user was alert 2) when the user wore goggles simulating inebriation 3) when the user was drowsy. For cognitive ability, I coded a game to test the subject's reflexes. I tested on subjects 1) while sober 2) while wearing "inebriated" goggles 3) while drowsy. To test for voice, I conducted two tests--one to measure slurring with the Google Speech Recognition API, and one to detect the volume of a person's voice using an algorithm integrating the AudioRecorder and FFT libraries. I tested on subjects 1) while awake 2) while drowsy. To test the impact of inebriation on voice, I used the ALC Corpus database, which provided voice samples of sober and intoxicated persons. I ran these samples through the same voice algorithm. Results I found that a person's balance was 3x as worse when "inebriated" than "sober." Also, 90% of my subjects deviated from their balance by 44% more than they did when drowsy. The reflex test showed that 75% of my subjects scored worse while "inebriated." 60% of my subjects scored worse on the reflex test while drowsy. From the slurring and volume tests, I found that pronunciation accuracy decreased by 40% for all subjects while volume decreased by about 20%. After testing samples from the ALC Corpus, I found that 66% of these samples showed a decrease in volume when inebriated. Conclusions/Discussion This experiment was successful in identifying the right biometrics for measuring inebriation/drowsiness through an intuitive smartphone app.

Awards (1)

• First Place

Competition history

• CSEF 2015 Behavioral & Social Sciences · Entry S0422

Resources

• CSEF project summary ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Behavioural and Social Sciences projects
• CSEF Behavioural and Social Sciences projects
• Every CSEF 2015 project
• CSEF First Place winners