In this paper we report progress on the use of novel data sets to calibrate and improve car-following models of highway traffic. These models describe vehicles as discrete entities moving in continuous time and space, and use plausible behavioural assumptions to derive systems of differential equations which must be solved for vehicles trajectories. To date, there has been little or no systematic attempt to fit these models to quantative microscopic (at the level of individual vehicles) highway data. Our recent work has involved the analysis of two such data sets:
Each of these data sets has been kindly provided by the Highways Agency.
The MIDAS (Motorway Incident Detection and Automatic Signalling) system consists of double inductance loops which record the lane number, time headway and wall-clock time of passing vehicles, as well as estimating their lengths and speeds. This data has been collected by TRL.
We have applied novel statistical techniques to the pattern matching of vehicle length measurements at different loops, and thus we are able to track individual vehicles down the motorway. In particular, we may:
However, presently MIDAS loops may not be used to estimate vehicles accelerations. We describe how image analysis and pattern matching techniques have been used to automatically follow vehicles across a sequence of aerial video frames. Projective transformations can then be applied to build full time series of vehicles displacements, from which headway, velocity, and acceleration information can be extracted.