Realtime Computer Vision-Based Accurate Vehicle Counting and Speed Estimation for Highways

Abstract

In Australian urban roads, pneumatic tubes are temporarily installed over roads to determine the road usage by vehicles. This is a relatively expensive process and the data cannot be obtained for about two weeks until a manual retrieval of data. This data is collected very randomly in order to determine the road usage in Australia. However, since the manual labor of installing such a device is very expensive in Australia, such deployments are rare and do not provide adequate information to the Road Maritime Services in Australia for future design and management of roads.

We have developed a highly accurate real-time computer vision-based system which relies on back ground subtraction, morphological operations and Gaussian filtering to track centroid of vehicles and accurately determine their speeds and count them. The system also sends speed alerts when congestion lowers the allowable speeds on highways. The real-time processing is achieved through optimization of our algorithms implemented using OpenCV. The code optimization has resulted in real-time operation of our system requiring no more processing power than that is available on a typical modern smartphone. The system is robust against typical lighting variations due to the movement of the sun and can maintain the accuracy in a drizzle or cloud movement making the system very practicable for deployment in Australian highways. In order to avoid the occlusion problem faced by many vision-based traffic monitoring systems, we have placed our cameras strategically above the highways with excellent results.