Devanshi Patel

Overview of the problem statement:

An IOT & Deep learning based DADAS system’s aim is to avoid nearly 70%* of road accidents in India due to drowsy or drunk drivers. It will also be ingrained with an innovative Analysis & Prediction mechanism for better traffic managements and future accident avoidance.
* Data taken from a report of Community against Drunken Driving (CADD)

My Role:

Brainstormed an idea for every given topics and possible solutions wrt physibility aspect.In depth analysis of each espect of the solution considering the technology espect and impact on human lifes including system design process.
Having worked on game design allowed me to take up the part of designing and presenting our idea in best form of articulation.
Once the idea was selected which is DADAS(Drunk and Drowsiness Alert System), we divided the overall system into main 4 parts,

1)Input & Preprocessing (DADAS IOT based Input Module)
2)Deep Learning (DADAS Cloud Server for Processing)
3)Alert Mechanism.
4)DADAS Analytics & Prediction System.

I took up the drunk(1st and 2 nd) part and I had to tech myself Machine Learning from scratch. For any Machine learning model construction foundation is dataset and unfortunatly we were not having any dataset for identification of drunk state/drunk people to train or neither did we have any pre-trained model. A solution to this was to take up our own sample dataset, I took up the part of collecting sample,data cleaning and making sure they are ready to be feed into ML(preprocessing). As we didn't have any pre-trained model we started with brute force approach in which I explored and taught myself how to train images over framework/libraries like caffee and tensorflow.
At the end after other teams mates involved in integrating and making an android application we were able to achieve 92% training accuracy using classification algorithms in drunk detection.