Development on an image processing system for traffic flow analysis

1.一种连续图像数字化自动处理系统，我们研制了一种从电视摄像机记录的视频图像中检测和跟踪运动车辆的自动处理系统。本系统采用两台计算机。1）视频图像数字化系统一台计算机用于存储视频图像。来自电视摄像机的视频信号由A/D转换器数字化为6位数。整个帧被划分为256**256个点，但只有帧的指定区域存储在计算机的存储器中。2）控制视频甲板的系统另一台计算机用于控制视频甲板。我们需要自动控制帧速率的值。因为最大帧速率受访问时间和传输速率的限制，大约为15帧/秒。2.应用需要分析冬季信号交叉口车辆微观行为。冬季路面湿滑，信号交叉口通行能力低于其他季节。本文应用一种新的测量系统对信号交叉口的单车二维测量进行了研究。我们可以在湿滑的路面上测量速度和加速度的值。但是...在一般条件下的实际应用中，还存在一些问题需要解决.车辆检测模型本文研究了一种在连续两帧视频中识别同一车辆的方法。神经网络模型之一。Congitron用于识别车辆图像。测试了认知器在车辆图像识别中的应用可能性。实验结果表明，训练后的模式可以识别，而未训练的模式不能识别。

1. An automatic processing system for digitizing sequential pictures.We developed an automatic processing system which detects and tracks moving vehicles from video image recorded by the TV camera. Two computers are used in this system.1) System for digitizing video imageOne computer is used to store video images. The video signal from a TV camera are digitized to a 6-bit number by an A/D converter. The entire frame is divided into 256**256 points, but only a specified region of the frame is stored in the computer's memory.2) System for control video deckAnother computer is used to control video deck. We need to control a value of frame rate automatically. Because, the maximum frame rate is limited by access time and transfer rate and it is about 15 frames/second.2. applicationIt is required to analyze the microscopic behavior of vehicles at signalized intersections during winter. In winter, road surface became very slippery, and capacity of signalized intersections in winter became less than that in other seasons. We applied a new system to study two-dimensional measurement of individual vehicle at signalized intersections. We could measure values of speed and acceleration on slippery road conditions. But. some problems remain to be solved for the practical applications in general conditions.3. Model for detecting vehiclesWe try to study a method to identify the same vehicle on two consecutive video frames. One of the Neural-Network-Model. Congitron was used to identify vehicle images. Application possibilities of cognitron for recognition of vehicle images were tested. Experimental results show that trained patterns could be recognized, but non-trained ones couldn't.