权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

Research on Traffic Flow Study Using Image Processing Method

基于图像处理方法的交通流研究

基本信息

批准号：
04650477
负责人：
KOIKE Hirotaka
金额：
$ 1.28万
依托单位：
Utsunomiya University
依托单位国家：
日本
项目类别：
Grant-in-Aid for General Scientific Research (C)
财政年份：
1992
资助国家：
日本
起止时间：
1992 至 1993
项目状态：
已结题

项目摘要

The purpose of this research is to develop a system to analyze traffic flow using image processing techniques. The system consists of video camera, VTR, NTSC decoder, image processor, minicomputer and page printer. Two methods were developed to extract vehicles from video image data. In both methods, image processing techniques were adopted to identify the location of cars on the video screen, and automatically digitize their coordinates.The first method is a subtract operation between two consecutive images, which leaves only the moving parts as residual images on the screen. Then the center of gravity of these residual images are computed. The problem with this method is the inaccuracy of center of gravity of a car when shadow exists or when a number of cars closely positioned.In the second method, a new approach was developed to further improve the accuracy of automatic recognition of cars using a procedure which is named as the differential growth method. The procedure works as follows. First, make the residual of subtracted images which shows the part of moving cars. Then applying the new method, the partial image of the car is led to grow to restore the original car shape. Once the original car image is restored, accurate center of gravity is computed and the trajectory of car movement is traced, or its speed is measured.Using this procedure, much more accurate movements of vehicles are automatically recognized. Case study at an intersection showed vehicle movement in the intersection fairly accurately. The application of this method seems promising in various traffic engineering surveys.

本研究的目的是开发一个利用图像处理技术分析交通流的系统。该系统由摄像机、录像机、解码器、图象处理机、小型计算机和页式打印机组成。提出了两种从视频图像数据中提取车辆的方法。这两种方法都采用图像处理技术来识别汽车在视频屏幕上的位置，并自动标定其坐标;第一种方法是对两幅连续图像进行相减运算，只留下运动部分作为屏幕上的残留图像;然后计算这些残差图像的重心。第二种方法针对阴影存在或多辆汽车密集时汽车重心不准确的问题，提出了一种新的方法，即微分增长法，进一步提高了汽车自动识别的准确性。该过程的工作原理如下。首先对减影后的图像进行残差处理，得到运动汽车的部分图像;然后应用新的方法，对汽车的局部图像进行生长，以恢复汽车的原始形状。在恢复原始车辆图像后，计算出车辆的准确重心，并跟踪车辆运动轨迹或测量其速度，从而自动识别出更准确的车辆运动。在交叉口的案例研究表明，车辆运动在交叉口相当准确。该方法在各种流量工程调查中的应用前景广阔。