基于车牌识别流数据的车辆伴随模式发现方法

发布时间：2018-04-17 09:07

本文选题：流式时空大数据 + 大数据分析　；参考：《软件学报》2017年06期

【摘要】：针对伴随车辆检测这一新兴的智能交通应用,在一种特殊的流式时空大数据——车牌识别流式大数据(ANPR)下,重新定义了Platoon伴随模式,提出PlatoonFinder算法,即时地在车牌识别数据流上挖掘Platoon伴随模式.主要贡献包括:第一,将Platoon伴随模式发现问题映射为数据流上的带有时空约束的频繁序列挖掘问题,与传统频繁序列挖掘算法仅考虑序列元素之间位置关系不同,该算法能够在频繁序列挖掘的过程中有效处理序列元素之间复杂的时空约束关系;第二,该算法融入了伪投影等性能优化技术,针对数据流的特点进行了性能优化,能够有效应对车牌识别流式大数据的速率和规模,从而实现车辆Platoon伴随模式的即时发现.通过在真实车牌识别数据集上的实验分析表明:PlatoonFinder算法的平均延时显著低于经典的Aprior和PrefixSpan等频繁模式挖掘算法,也低于真实情况下交通摄像头的车牌识别最小时间间隔.因此,所提出的算法可以有效地发现伴随车辆组及其移动模式.
[Abstract]:Aiming at the new intelligent traffic application of accompanied vehicle detection, this paper redefines the Platoon adjoint pattern under a special flow space-time big data-vehicle license plate recognition flow big data / ANPRA, and proposes the PlatoonFinder algorithm.Platoon concomitant patterns are instantly mined on the license plate recognition data stream.The main contributions are as follows: first, the Platoon adjoint pattern discovery problem is mapped to the frequent sequence mining problem with temporal and spatial constraints on the data stream, which is different from the traditional frequent sequence mining algorithm, which only considers the location relationship between sequence elements.The algorithm can effectively deal with the complex temporal and spatial constraints between the sequence elements in the process of frequent sequence mining. Secondly, the algorithm integrates pseudo projection and other performance optimization techniques to optimize the performance of the data flow.It can effectively deal with the speed and scale of vehicle license plate recognition flow big data, so as to realize the real-time discovery of vehicle Platoon concomitant pattern.The experimental results on the real license plate recognition data set show that the average delay of the weighted Platoon Finder algorithm is significantly lower than that of the classical frequent pattern mining algorithms such as Aprior and PrefixSpan, and also lower than the minimum time interval of the real traffic camera license plate recognition.Therefore, the proposed algorithm can effectively discover the associated vehicle groups and their moving patterns.
【作者单位】：天津大学计算机科学与技术学院;大规模流数据集成与分析技术北京市重点实验室(北方工业大学);北方工业大学云计算研究中心;
【基金】：国家自然科学基金(61672042) 北京市市委组织部、北京市优秀人才培养资助,青年骨干个人项目北方工业大学“人才强校计划”青年拔尖人才培育计划~~
【分类号】：TP311.13;U495

【相似文献】