Real Time Accident Prediction and Related Congestion Control Using Spark Streaming in an AWS EMR cluster

This paper focuses on real-time cloud based analytics of live video feeds from the cameras of self-driven autonomous vehicles using the Spark framework on Amazon's Elastic Mapreduce (EMR). We use deep-learning methodologies for real-time object detection on the streamed images, to classify and predict traffic incidences, leading to subsequent congestion control. Results on a benchmark application: traffic congestion aware navigation using 10 self-driving vehicles with their own camera feeds as they drive around in Manhattan, show a 60% improvement in performance on the AWS-EMR based Spark framework, when compared to cloud processing on single instance of EC2 server on the AWS.