The aim of this project is to produce a low-cost, easy to use, automatic traffic data collection system, using video recordings played back through a video digitiser to a microcomputer. The system, referred to as TULIP (Traffic analysis Using Low Cost Image Processing), currently counts traffic and measures headway and speed, all in real time. The equipment for the project consists of an IBM PC-AT microcomputer which acts as the host to an image processing subsystem, which is a commercially available device, the FG100-AT, produced by Imaging Technology. The image processor can accept video signals from any commercially available video recorder or camera, digitise them for use by a computer and store the images in frame memory. This is all done in real- time, namely 40 msec, the time it takes to transmit one full TV frame. In addition, the FG100-AT also contains an input look up table which can be programmed to perform some useful primitive processing of the image, also in real-time. Real-time counting has been performed with this system, utilising all of the 25 frames per second of video data. Two observation windows are projected into the scene by the system software and processing is concentrated on the pixels in these areas. This is necessary because of the vast amount of data present in a video image, the majority of which is of no interest because no vehicles ever appear. A simple filter is applied to the pixel intensities to highlight vehicles in the scene. A temporal filter is then applied to reduce the effects of random noise. Vehicles have been counted as they passed between the two observation windows with an accuracy of between 1.1 and 8.6 per cent under good lighting conditions. The speed of the vehicles can also be gauged,in real time, if the observation windows are placed a known distance apart. The software is presently being refined to improve the performance in poor and variable lighting conditions.
Abstract