基于云计算的出租车异常行为探测研究

发布时间：2018-07-04 20:19

本文选题：异常轨迹检测 + 地图网格化　；参考：《武汉理工大学》2015年硕士论文

【摘要】：随着传感技术、通讯技术、储存技术和计算能力的发展,越来越多出租车装配了GPS传感仪,在日常运营中产生大量位置数据,为我们提供了很好的机会去分析和挖掘有价值的信息。本文主要将数据应用于探测出租车异常行为,目标是自动识别车辆异常行驶轨迹,判断司机是否存在故意绕路行为。既能保障乘客利益也有助于规范出租车服务,具有现实意义。本文主要研究工作如下:(1)为实现检测出租车异常轨迹的目标,本文先给出轨迹等相关定义,设计了检测系统总体框架,按模块说明各节点作用,并从离线和在线处理阶段分析了数据处理流程。(2)为解决轨迹网格化后存在的不连续问题,本文提出AE-AUG(Augmented method of angle and edge)轨迹补全算法,可快速求出一条通路连接两不相邻网格。(3)对异常轨迹检测核心问题,本文提出s-iBOAT(iBOAT with State)算法,该算法通过为轨迹加入最新检测点位置下标,改进基于孤立特性的在线异常轨迹检测算法iBOAT(Isolation-Based Online Anomalous Trajectory Detection),简化查找相似轨迹处理步骤,提升算法效率。(4)利用Hadoop平台处理出租车GPS记录生成历史轨迹数据,结合Bing Maps Tile System中的地图网格算法和本文提出的AE-AUG、s-iBOAT算法,实现了一个基于Web前端技术的异常轨迹检测系统。通过检测系统测试了s-iBOAT算法异常子轨迹识别效果,实验结果良好符合理论分析。对相同起终点所有运营轨迹进行检测,从总体检测情况分析两种异常轨迹形成的原因。其一是部分出租车司机经验丰富,对该区域熟悉,选取了少数的便捷路径导致识别异常。其二是司机为了攒取更多的运营收益,载客时故意绕远路导致异常。研究异常判别阈值和网格大小对检测灵敏度、误判率、准确率的影响,在测试实验条件下得出两者最佳取值为0.1和153米。此外利用异常轨迹长度与总体历史轨迹集平均长度对比情况修正检测结果,实验表明可提高检测准度,更适用于探测现实出租车绕路行为。对比检测算法改进前后的执行效果,结果显示s-iBOAT能够保持异常子轨迹识别效果基本不变、整体轨迹检测准度相同的情况下提高运行速度,减少响应时间。
[Abstract]:With the development of sensing technology, communication technology, storage technology and computing ability, more and more taxis are equipped with GPS sensor, which produces a lot of position data in daily operation. It provides us with a good opportunity to analyze and mine valuable information. This paper mainly applies the data to detect the abnormal behavior of the taxi. The goal is to automatically identify the abnormal track of the vehicle and judge whether the driver has the behavior of deliberately detour. It is of practical significance to protect the interests of passengers as well as to standardize taxi service. The main research work of this paper is as follows: (1) in order to realize the goal of detecting the abnormal track of taxi, this paper first gives the definition of track and other related definitions, designs the overall framework of the detection system, and explains the role of each node according to the module. The data processing flow is analyzed from off-line and on-line processing stages. (2) in order to solve the discontinuous problem of trajectory gridding, an AE-AUG (Augmented method of angle and edge) trajectory complement algorithm is proposed in this paper. Two nonadjacent meshes can be quickly solved by a single path. (3) for the core problem of abnormal trajectory detection, this paper proposes s-iBOAT (iBOAT with State) algorithm, which subscribes the position of the latest detection point to the trajectory. The algorithm iBOAT (Isolation-Based online Anomalous Trajectory Detection) is improved to simplify the process of finding similar tracks and improve the efficiency of the algorithm. (4) the Hadoop platform is used to process taxi GPS records to generate historical track data. Combined with the map grid algorithm in Bing Maps Tile system and the AE-AUGCS-iBOAT algorithm proposed in this paper, an anomaly track detection system based on Web front-end technology is implemented. The detection system is used to test the detection effect of s-iBOAT algorithm. The experimental results are in good agreement with the theoretical analysis. The causes of the formation of the two abnormal trajectories are analyzed from the overall detection of all the operation tracks of the same starting and ending points. One is that some taxi drivers are experienced, familiar with the area, and select a few convenient paths to identify anomalies. The other is the driver in order to save more operating income, when carrying passengers deliberately detour caused abnormal. The effects of abnormal threshold and mesh size on detection sensitivity, error rate and accuracy are studied. The optimum values of them are 0.1 and 153 meters under the test conditions. In addition, by comparing the abnormal track length with the average length of the total historical track set, the experimental results show that the detection accuracy can be improved, and it is more suitable to detect the actual taxi detour behavior. The results show that s-iBOAT can keep the recognition effect of abnormal sub-trajectory unchanged and improve the running speed and reduce the response time when the whole track detection accuracy is the same.
【学位授予单位】：武汉理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495

【参考文献】