Statistical Estimation from Videos of Freeway Traffic

University Of California-Berkeley, Berkeley CA

Investigators

Abstract

The proposed research focuses on the estimation of velocity, density, and flow fields from video recordings of freeway traffic, using data from the Berkeley Highway laboratory, which has recorded thousands of hours of traffic flow on freeway I-80. The investigators are particularly concerned with recordings which are not sufficiently well resolved to allow unambiguous vehicle identification and tracking. They propose to investigate two broad classes of methods. The first, building on an initially encouraging approach, is based upon local semiparametric statistical modelling of the velocity field. The second is based upon simulation using microscopic models. Both of these present challenges for the development of statistical methodology and theory. Broader impacts include: the development of effective, practical methods for video surveillance of traffic, which would be of widespread utility; the compilation of a large empirical body of velocity fields, which will be a valuable resource for researchers in traffic flow theory; extension of the methodology to the more general problem of estimating optical flow in two dimensions, which has widespread applications in computer vision; and the involvement of undergraduates and graduate students in research.