面向套牌甄别的流式计算系统

Stream computing system for monitoring copy plate vehicles

套牌车的甄别具有时效性约束。针对现有计算检测方法中所出现的精度低、响应慢等局限,提出了一种基于实时车牌识别(ANPR)数据流的套牌车流式并行检测方法,设计了基于路段阈值表和时间滑动窗口的套牌计算模型,能够实时地甄别出交通数据流中的套牌嫌疑车。在Storm环境下,利用某市真实交通数据集模拟成实时交通流数据进行实验和评估,实验结果表明计算的准确率达到98.7%,并且一条车牌识别数据的处理时间为毫秒级。最后,在该计算模型基础上实现了套牌车稽查防控系统,能实时甄别并展现出当前时刻城市交通网中出现的所有套牌嫌疑车。

The screening of the copy plate vehicles has timeliness, and the existing detection approaches for copy plate vehicles have slow response and low efficiency. In order to improve the real-time response ability, a new parallel detection approach, called stream computing, based on real-time Automatic Number Plate Recognition （ANPR） data stream, was proposed. To deal with the traffic information of the road on time, and plate vehicles could be timely feedback and controlled, a stream calculation model was implemented by using the threshold table of road travel time and the time sliding window, which could access real-time traffic data stream to calculate copy plate vehicles. On the platform of Storm, this system was designed and implemented. The calculation model was verified on a real-time data stream which was simulated by the real ANPR dataset of a city. The experimental results prove that a piece of license plate recognition data can be dealt with in milliseconds from the time of arrival to the calculation completion, also, the calculation accuracy is 98.7%. Finally, a display system for copy vehicles was developed based on this calculation model, which can show the copy plate vehicles from the road network on the current moment.