Direct Methods of Determining Traffic Stream Characteristics by Definition

There are generally two approaches to derive traffic stream characteristic s (i.e. flow, space mean speed, and density): indirect approach which derives traffic stream characteristics by estimation from correlated variable(s) and direct approach which determines traffic stream characteristics by definition from field observations. A method of the former approach inevitably involves estimation error while a method of the latter provides “the ground truth”. This paper examines and contrasts three direct methods, namely the HCM method, the x-t method, and the n-t method. The HCM method determines traffic stream characteristics based on their definition adopted by Highway Capacity Manual (HCM). The x-t method determines traffic stream characteristics based on their definition proposed by Edie in generalized terms. The x-t method is so named because it works on vehicle trajectories in space-time (x-t) domain. The n-t method also determines tr affic stream characteristics ba sed on Edie’s definition, but the method works on cumulative ve hicle counts in cumulative number-time (n-t) domain. In terms of accuracy, the x-t method and the n-t method are almost equivalent and both are more accurate than the HCM method. In terms of the fundamental relationship among flow, speed, and density, the x-t method and the n-t method guarantee the relationship while the HCM method does not. In terms of the working mode, the HCM method and the n-t method are capable of working on-line/real-time while the x-t method typically works off-l ine. In terms of data sources, the n-t method uses as input point sensor data while the x-t method and the HCM method require aerial photograph data or a combination with others. In summary, the n-t method provides a promising means for traffic management centers (TMC) to accurately determine traffic stream ch aracteristics by definition on-line/real-time from data collected by ITS traffic surveillance systems th at predominantly consist of point sensors