Statistical Estimation from Videos of Freeway Traffic

这项研究的重点是利用伯克利高速公路实验室记录的I-80高速公路上数千小时的交通流量数据，从高速公路交通的视频记录中估计速度、密度和流场。调查人员特别关注的是那些分辨率不够高，无法明确识别和跟踪车辆的录音。他们建议研究两大类方法。第一种方法是基于速度场的局部半参数统计建模，这是一种最初令人鼓舞的方法。第二种是基于微观模型的模拟。这两者都对统计方法和理论的发展提出了挑战。更广泛的影响包括：开发有效、实用的交通视频监控方法，这将具有广泛的实用性；积累大量的速度场经验，为交通流理论研究提供宝贵的资源；将该方法扩展到更一般的二维光流估计问题，该问题在计算机视觉中有广泛的应用；以及本科生和研究生参与研究。

The proposed research focuses on the estimation of velocity, density, andflow fields from video recordings of freeway traffic, using datafrom the Berkeley Highway laboratory, which has recorded thousandsof hours of traffic flow on freeway I-80. The investigators are particularlyconcerned with recordings which are not sufficiently well resolvedto allow unambiguous vehicle identification and tracking.They propose to investigate two broad classes of methods. Thefirst, building on an initially encouraging approach, is basedupon local semiparametric statistical modelling of the velocityfield. The second is based upon simulation using microscopicmodels. Both of these present challenges for the development ofstatistical methodology and theory.Broader impacts include: the development of effective, practicalmethods for video surveillance of traffic, which would be ofwidespread utility; the compilation of a large empirical body ofvelocity fields, which will be a valuable resource for researchersin traffic flow theory; extension of the methodology to the moregeneral problem of estimating optical flow in two dimensions,which has widespread applications in computer vision; and theinvolvement of undergraduates and graduate students in research.