Traffic Flow Parameter Measurement And Model Research Based On Traffic Video Data

Nowadays traffic congestion has become a serious problem. While road widening is becoming less and less possible, researchers have drawn their attention to using the existing infrastructure more efficiently. Traffic flow models are important tools to investigate complex traffic phenomena and used to establish traffic control systems. However, though a lot of improvement has been made in traditional traffic flow models and new theories keep coming up (Greenshields,1934; Lighthill and Whitham, 1955; Greenberg,1959; Underwood,1961; Payne,1979; Wu,1994; Daganzo,1995, Jiang.2002, Xue 2003), level of congestion seems to be ever-growing. Besides, there are some specific empirical observations that many traffic flow models cannot reproduce. Such situation has caused traffic flow research to focus on empirical analysis. On one hand, the formation of a traffic flow model requires massive empirical observation as well as in-depth investigation of real road traffic, in which case empirical data are indispensable to identifying parameters when the corresponding model is utilized. On the other hand, traffic flow modeling is now at a point where researchers are trying to identify the best of all models. And basically this is to be done through testing models against empirical data.Hence, not only is there a great need for empirical data, but also one for empirical data-based models to better describe the real mechanisms and properties of traffic congestion. It is of primary importance to study the spatiotemporal vehicle behaviors by analyzing micro-data of individual vehicles. Early traffic flow empirical researches date back to 1990s when such researchers as Kerner (1993,1994,1996,1998,2000), Helbing (1998,2001,2007) and Knospe (2004) used data obtained by loop detectors buried underground to put forth non-linear velocity-density traffic flow models and a series of new complex traffic theories, among which Kerner's famous "three-phase traffic theory" was mainly based on the study of traffic on the German freeway A5 between 1996 and 2001. It must be pointed out that the direct output of loop detectors is usually velocity and total number of vehicles passing in a certain period of time. And before these data are used in traffic flow modeling, a corresponding density data must be averaged out to match each velocity data. During this process, simultaneity is lost between density data and velocity data.However, traffic video is found to be able to provide simultaneous micro-data of vehicles and thus solve the issue. By image-processing methods, the headway distance of an individual vehicle can be calculated and then used as a density data for modeling after being inversed, while velocity data of the same vehicle can be gotten from its displacement during two consecutive video frames. In this sense, velocity and headway distance have better simultaneous characteristics than data directly output from loop detectors.In this paper, a one-dimensional grey-level vehicle detection method is applied as a critical component of a video-based system for traffic parameter measurement. This method restricts the detection area from two-dimension to one -dimension, so it has the natural advantage of less computation and better efficiency. The system acquires a large amount of velocity and headway distance data as main parameters to investigate the traffic flow of urban expressways.Based on videos of traffic flow at several urban freeway sections of some typical cities in China, a large database which contains 210,920 "vehicle speed-headway distance" data pairs measured from the real traffic flow is obtained. The measured data show consistent with the fundamental nature of traffic flow since their average speeds increase as the headway distances increase. The average speed usually decreases from ultra-lane to exchange-lane, but there exist "density inversion" cases when the road sections are nearly exits. It is found from the flow-density fundamental diagrams drawn by the measured data that there is a type of basic graph in which nonlinear characteristics show in almost the whole density area. Over 7% "high speed car-following" vehicles are found in the small headway areas.This paper also discusses one-order and two-order traffic flow models, especially 1-d pipe flow model. Based on a first integral of 1-d pipe flow model, an analytical solution is designed. Besides, a compact difference scheme with two free parameters is created.