基于车牌流数据的伴随车发现方法研究

发布时间：2018-09-14 18:29

【摘要】：随着社会经济的快速发展,机动车越来越多,人们出行愈加频繁。在中国以北京、上海、深圳为代表的大城市,智能交通管理监控系统每天都会存储海量的机动车出行数据,包括ANPR数据、GPS数据等。充分利用这些交通数据挖掘有效的信息,解决以伴随车分析为例的分析性型交通问题,一直以来是智能交通领域的研究热点。伴随车分析作为分析性类型的交通问题之一,其典型的应用场景是"犯罪团伙驾车协同作案"。此类案件的理想解决方案是在犯罪团伙逃逸过程中就能及时发现嫌疑车辆,通知公安干警前往拦截,实时性要求较高。近年来基于流数据的研究逐渐成为大数据研究的一个趋势。本文基于随时间变化的车牌识别流数据,通过实时分析多辆犯罪嫌疑车辆在一段时间内的关系,提出了一种并行PFID算法的伴随车辆发现方法。利用成熟的分布式流数据处理框架Spark Streaming实现了该算法,在秒级响应时间内找到伴随车辆组,达到及时预警效果,便于公安部门及时处理。本文的主要工作如下:(1)提出了一种并行PFID算法的伴随车辆发现方法。根据伴随车辆组的定义、车牌识别数据的数据格式以及现有的数据模型技术,经过分析后,采用关联规则挖掘相关算法来解决本问题。PFID算法采用了关联规则Eclat算法的思想,基于车牌流数据进行频繁项集的挖掘,从而发现伴随车辆组。(2)利用了分布式流数据处理框架Spark Streaming实现了该PFID算法。在云环境下的Spark集群中进行实验,从内存和响应时间两方面进行对比。实验结果表明,该方法在内存消耗和响应时间上都有较好的效果,克服单一机器下程序运行内存不足等问题,较快地发现伴随车辆组。(3)构造了伴随车辆组发现的原型系统。为了更好地呈现实验结果,构造了原型系统,将结果以图表、地图等可视化呈现,使实验结果更形象生动。(4)实现了伴随车辆组服务化。为了便于第三方使用伴随车辆组结果数据,本文将通过实验处理后的伴随车辆组结果数据以REST风格的Web API对外提供,对外提供的数据格式包括Text、XML、JSON等。
[Abstract]:With the rapid development of social economy, more and more motor vehicles, people travel more frequently. In China's big cities, such as Beijing, Shanghai and Shenzhen, the intelligent traffic management and monitoring system stores huge amounts of motor vehicle travel data every day, including ANPR data and so on. Using these traffic data to mine effective information to solve the analytical traffic problem, which takes the analysis of accompanying vehicles as an example, has always been the research hotspot in the field of intelligent transportation. As one of the analytical traffic problems, the typical application scene of accompanying vehicle analysis is "gang driving Synergistic Crime". The ideal solution for this kind of case is to find the suspected vehicle in time during the escape of the criminal gang, to notify the police to stop the case, and to have a high real-time requirement. In recent years, the research based on stream data has gradually become a trend of big data research. Based on the time-varying license plate recognition stream data, this paper presents a parallel PFID algorithm based on the real-time analysis of the relationships of several suspected vehicles over a period of time. The algorithm is implemented by using the mature distributed stream data processing framework (Spark Streaming), which finds the accompanying vehicle group in the second response time, and achieves the effect of timely warning, which is convenient for the public security department to deal with the problem in time. The main work of this paper is as follows: (1) A parallel PFID algorithm for adjoint vehicle discovery is proposed. According to the definition of associated vehicle group, the data format of license plate recognition data and the existing data model technology, after analysis, the association rule mining algorithm is adopted to solve the problem. PFID algorithm adopts the idea of association rule Eclat algorithm. The frequent itemsets are mined based on the license plate stream data, and the associated vehicle groups are found. (2) the PFID algorithm is implemented by using the distributed stream data processing framework (Spark Streaming). The experiment is carried out in Spark cluster in cloud environment, and the memory and response time are compared. The experimental results show that the proposed method has good performance in memory consumption and response time. It overcomes the problems of running memory in a single machine and finds the associated vehicle group quickly. (3) A prototype system with vehicle group discovery is constructed. In order to better present the experimental results, a prototype system is constructed, and the results are visualized as graphs and maps. (4) Service-oriented accompanying vehicle groups are implemented. In order to facilitate the third party to use the accompanying vehicle group result data, the result data of the accompanying vehicle group will be provided by REST style Web API through the experiment. The external data format includes Text,XML,JSON and so on.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495;TP311.13

【参考文献】