Use of Image Analysis to Estimate Average Stopped Delays per Vehicle at Signalized Intersections
Use of Image Analysis to Estimate Average Stopped Delays per Vehicle at Signalized Intersections
复制标题
DOI：
10.3141/1776-14
复制
发表时间：
2001
影响因子：
1.7
通讯作者：
M. Saito;J. Walker;A. Zundel
中科院分区：
文献类型：
--
作者：
M. Saito;J. Walker;A. Zundel
平均停止延迟被用来确定信号交叉口的服务水平。这项工作大多是通过人工现场观测方法、分析模型和仿真软件在线下完成的。由于常规方法需要大量的人力和成本，因此对信号交叉口的目标值进行动态评估一直是困难的。然而，许多交通运营中心希望在发生拥堵时能够协调地面街道和高速公路的管理。结果表明，图像分析可以动态确定平均停止延迟。提出了间隙法和运动法两种方法，并与ITE的场延迟观测方法进行了比较。交通软件集成系统软件用于创建排队车辆的图像。为了消除视差问题，我们做了两个简单的假设：从与人行道垂直的方向观察排队的车辆，并且所有的视线都平行于摄像机的观察方向。目前，算法很简单，但它们产生了有希望的结果。用1 ~ 20秒的不同分析时间间隔估计停止延迟。当时间间隔约为15 s时，间隙法得到了较好的结果；运动法在时间间隔为1或2 s时效果较好。人工观测方法得到了与模拟软件相似的较好的估计，特别是当时间间隔在10秒左右或更短时。
Average stopped delay has been used to determine the level of service (LOS) of signalized intersections. This work has mostly been done offline by using manual field observation methods, analytical models, and simulation software. Dynamic evaluation of the LOS of signalized intersections has been difficult because the use of conventional methods requires a large amount of labor and cost if such analyses were conducted periodically. However, many traffic operations centers wish to be able to coordinate surface street and freeway management when congestion occurs. It is demonstrated that image analysis can be used to dynamically determine average stopped delay. Two methods—the gap method and the motion method—were developed and compared with ITE’s field delay observation method. Traffic Software Integrated System software was used to create images of queued vehicles. To eliminate parallax problems, two simple assumptions were made: the queued vehicles were viewed from the direction normal to the pavement, and all lines of sight were parallel to the camera’s view direction. At present, the algorithms are simple, but they produced promising results. Various analysis time intervals from 1 to 20 s were used to estimate stopped delay. The gap method gave a good result when the time interval was about 15 s; the motion method produced a good result when the time interval was 1 or 2 s. The manual observation method resulted in a good estimate similar to that which the simulation software produced, especially when the time interval was about 10 s or less.