68 Extracting Traffic Parameters at Intersections Through Computer Vision Proceedings of the Technical Sessions, 27 (2011) 68-75 Institute of Physics – Sri Lanka Extracting Traffic Parameters at Intersections Through Computer Vision K.D. Peiris and D.U.J. Sonnadara Department of Physics, University of Colombo ABSTRACT A vision-based system for extracting important traffic parameters such as vehicle count, density, type and location at a three-way junction is presented. A single digital camera is used to acquire the video clip at the junction. In order to cover a wide area of the junction and to minimize the occlusion of vehicles, the camera is placed on top of a building near the junction. The accuracy of the system in vehicle counting is 91% and that of vehicle classification is 89% for small, medium and large vehicles. The errors in vehicle counting and classification occur mainly due to the occlusion of the moving vehicles. Errors also increase when the distance from the camera to the moving vehicle is high. Many of the problems in real-time implementation can be avoided to a great extent by using a multi camera system that can cover a large field of view. 1. INTRODUCTION Information on traffic is important in the planning, maintenance, and control of any modern transport system. To resolve traffic congestion and to make good use of roads, solutions are often sought through traffic signal control systems and traffic information services. Traffic surveillance technologies play an essential role in incident detection, traffic management, and travel time estimation. There are two basic types of traffic surveillance systems: road-based and vehicle-based. Road-based detection systems like inductive loop detectors have been a principal element of freeway surveillance and incident detection for many years. Similarly, video image detection and other roadside detection technologies have been used extensively to measure high-volume traffic conditions. However, advances in vehicle sensors and detection algorithms have given transportation authorities the opportunity to implement or enhance vehicle-based surveillance systems. Therefore, extensive efforts have been devoted to video-based measurement and analysis of traffic flow, which will greatly benefit current and future applications, including traffic control and analysis, violation detection, and vehicle identification. Today, vision-based traffic monitoring systems are necessary to improve the traffic flow especially in heavy traffic sections. Vision-based systems are capable of providing dynamic traffic information. A video sensor is one of the most important sensors to obtain information at many road points. Vision-based camera systems are more sophisticated and powerful than those based on spot sensors, i.e., loop detectors and pneumatic sensors, since the information content associated with image sequences allows vehicle tracking and classification [1]. The application of