Vision Systems for Analysis of Congested Traffic

For long-span bridges, congested traffic forms the governing traffic load condition. This paper reviews the current state of the art and the deficiencies in the data upon which the modelling of congestion is typically based. It is proposed that a vision system be adopted to collect statistical information for the parameters required to properly calibrate congestion models. Conventional methods for the detection of vehicle features in images are described, applied to sample images, and criticised. It is concluded that vision-based methods are a viable tool for gathering inter-vehicle gap data, but that existing methods of wheel detection are not sufficient. Therefore the development of a new technique for wheel detection is recommended and some suitable attributes identified.