4. Business Problem
● Drowsy driving is an unaddressed issue:
○ Over a 30 day period, 1/25 adult drivers reported having fallen asleep while driving (CDC)
○ Each year, drowsy driving accounts for about 100,000 crashes, 71,000 injuries, and 1,550
fatalities, according to the National Safety Council (NSC). Drowsy driving contributes to an
estimated 9.5% of all crashes, according to AAA.
○ Average 60% increase in insurance premium after an accident (US).
○ Car crashes accounted for 31 percent of all economic costs ($76.1 billion) of all motor vehicle
crashes (US).
● Currently, there is no universal mechanism for automatically waking up drivers when they feel tired.
5. Solution
● A retrofitting system that can enable more cars with best of what the world of AI has to offer for
road safety
● To detect a distracted driver.
● Sound alarm to wake / alert driver.
● Universal product:
○ mobile application
○ dedicated hardware
6. ETL and Storage
Server side:
● Initially data sourced from public sources.
● Cloud based database holds the bulk of data (to be hosted on AWS S3)
● Database used to train model on the cloud using databricks.
Client side:
● Transformation of live feed (greyscaling / cascade classification)
● Machine learning history stored locally for speedy access (175 MB images size to ~30 MB of
local storage required)
9. Data Cleaning and Preparation
Passive data preprocessing:
● Dimming
● Resolution
● Resizing
● Rotation
● Width-height shift
● Vertical-horizontal flip*
* changes from original plan
10. Data Cleaning and Preparation
Active data preprocessing (on the fly):
● Frames are turned grey (greyscaling)
● Cascade classification
20. Future Scope for improvement
● Fully working proof of concept was created.
● Future work can include adaptive learning based on driver behavior, which may require
client side to talk with a server to also detect yawning.
● Possibilities to explore options to integrate the application on the web using Django / Flask,
include compatibility with the phone’s camera as well.
● Expanded dataset containings tens of thousands of images.
● We also tried to implement a macOS native app.
21. ● The Economic and Societal Impact Of Motor Vehicle Crashes, 2010 (Revised), NHTSA
● Drowsy Driving: Asleep at the Wheel, CDC
● Rapid Object Detection Using a Boosted Cascade of Simple Features, Viola P Jones, M. Jones
● Images: Shutterstock, Google images
References