Camera-based Smart Traffic State Detection in India using Deep Learning Models

In the age of smart cities, it is more than necessary that a real-time road accident detection system exists that can automatically identify the vehicle, predict its level of damage, and notify emergency services for quick perusal. In this paper, we build an intelligent traffic monitoring system using state-of the-art deep learning models that aims to detect traffic accidents, burning vehicles, and traffic state (congestion/sparse traffic) using a single model for real-time monitoring purpose. Web-crawling was done to create a custom-built dataset for Indian traffic conditions. The developed models were found to perform decently in Indian traffic conditions with more than 90% accuracy showing the efficacy of the models in Indian conditions.