Driving the Last Mile: Characterizing and Understanding Distracted Driving Posts on Social Networks

Hemank Lamba, Shashank Srikanth *, Dheeraj Reddy *, Shwetanshu Singh, Karandeep Juneja, Ponnurangam Kumaraguru

Drinking & Driving	Passenger Distraction	Distracted Driving on Bikes

This repository contains code and data required to reproduce the results of "Driving the Last Mile: Characterizing and Understanding Distracted Driving Posts on Social Networks". (Paper link)

Introduction

Our work deals with identifying distracted driving content on social media (Snapchat) using computer vision methods and analyzing the data for spatio-temporal patterns and charecterizing the users demographics.

A few examples of distracted driving content on Snapchat are given in the figure above.

Installation

The code has been tested with python3 and PyTorch 0.4.1. The codebase can also support PyTorch 1.0 with slight modifications.

virtualenv -p python3.5 venv
source venv/bin/activate
pip install -r requirements.txt

Running the code

The codebase is divided into the following major portions:

1. Annotation Portal
2. Scraper
3. Data collection Pipeline
4. Classification
5. Analysis

Pre-trained Models and Scraper code

All the pretrained models and the Snapchat scraper code are available on request via email at pk [at] iiitd [dot] ac [dot] in.

Project Page

For additional details, plots, and discussions, refer to the project page here.

Citation

If you are using this code or the pre-trained models, please cite our paper as follows:

@article{lambadistracteddriving,
  author={Hemank Lamba, Shashank Srikanth, Dheeraj Reddy Pailla, Shwetanshu Singh, Karandeep Juneja, and Ponnurangam Kumaraguru},
  title={Driving the Last Mile: Characterizing and Understanding Distracted Driving Posts on Social Networks},
  booktitle={Proceedings of the International AAAI Conference on Web and Social Media (ICWSM) 2020},
  year={2020},
}