SteerSecure
A deep learning web-application that is made to ensure your safety as a driver.
Created on 22nd November 2020
•
SteerSecure
A deep learning web-application that is made to ensure your safety as a driver.
The problem SteerSecure solves
SteerSecure is a driver-safety app built using machine learning libraries such as Tensorflow and OpenCV, and scripted and deployed to the web using Streamlit. Our aim with SteerSecure is to make vehicular travel safer. Our two primary web-applications are:
- Distraction Detection
- Drowsiness Detection
The data is very clear: more than 30% of accidents are caused due to fatigue caused by long, nonstop hours of driving and sleep-deprivation. A large majority of accidents are also caused by distracted drivers: texting, speaking to passengers, talking on the phone, etc. Such accidents cause more than 400 deaths every day in India. Not only is it a tragic loss of life, but it also costs our economy upwards of 1 trillion rupees every year.
Our goal, very simply, is to curb this loss of life and property as best we can. SteerSecure aims to discourage distracted drivers and encourage drowsy drivers to take rest and pull over before continuing on the journey later.
Challenges we ran into
- The real-time predictions of the distraction model were inaccurate in the web-app while being perfectly fine in the Python notebook.
- We faced a lot of issues with deployment as well, involving different Conda environments and due to the size of our Github repo.
- Training the VGG16 model was quite an intensive task for our notebooks. The final model.h5 file turned out to be very heavy and we had to carefully install all other packages to try and not overshoot the resource limits.