IMAGE PROCESSING APPLIED TO TRAFFIC
Roger C. Waterfall, K W Dickinson · Traffic engineering & control · 1984
This paper examines the development of an automated image-analysis system for recording and analysing road traffic movements, describes its operation and outlines current limitations. Methods of detecting vehicles in a video scene and of reducing the effects of noise in the image are discussed. The performance and suitability of CCTV, infrared and solid-state cameras are considered. A pipeline preprocessor, which allows video-record to be compressed and stored for subsequent analysis, was designed and built. Field trials with the pipeline system showed that the computer identified 94 percent of vehicles observed by a human operator. With current technology the use of image-analysis systems for automatic collection of traffic data is limited to relatively simple procedures such as vehicle counts, speed, length and lane occupancy information as traffic crosses a datum line on the highway. The paper concludes by indicating the degree of improvement necessary before a flexible system is available with sufficient computing power to perform image analysis on traffic data and abstract the required results. See also TRIS No. 393258. (TRRL)