Drunk User Interfaces: Determining Blood Alcohol Level through Everyday Smartphone Tasks

on YouTube ↗23m

" summary

A smartphone system called the DUI app implements five tasks (typing, swiping a 3x3 passcode, PPG heart-rate hold, simple and choice reaction task) plus accelerometer/gyroscope features and trains a random-forest regression against breathalyzer readings. Evaluated on 14 people in a multi-day controlled study, it achieves a Pearson r = 0.96 and an absolute mean error of 0.005% ± 0.007%, with sensitivity 93.9% and specificity 82.3%, but it requires per-user calibration and quiet conditions to reach those numbers.

AI-written summary · about

" chapters (12)

1. 0:00Introduction to drunk user interfaces
2. 1:49Motivation and project goals
3. 3:56Related work and context
4. 5:23Defining drunk user interfaces
5. 6:26Task selection and design
6. 9:36Typing and reaction tasks explained
7. 11:45Study design and IRB challenges
8. 14:37Initial study results and issues
9. 16:23Multi-day study approach
10. 17:47Calibration and evaluation methods
11. 19:51Limitations and future work
12. 20:42Q&A session