Scalable Traffic Video Analytics using Hadoop MapReduce

摘要

Road traffic video analytics aims at using a number of techniques to achieve better traffic and road safety, control congestion and provide immediate care for accident victims. In this paper, we propose a near real-time traffic analytics system which can automatically detect road accidents from live video streams. The system alerts nearby hospitals and highway rescue teams when accidents occur. It also detects road congestion and broadcasts alternative route information to relevant commuters. We have implemented the system using Apache Hadoop. Analysis results are stored in Hive, which is a data warehouse built on top of Hadoop. This data includes overall traffic speed, traffic volume, and individual vehicle speed. Hive provides data summarization, query, and analysis. We have deployed and tested the system on a cluster computer.