摘要：

目前基于视频的异常行为检测主要针对单车受限场景,难以对运输全过程进行监控。而GPS轨迹分析也以对单车进行先验阈值判断为主,缺乏数据深层次分析与信息挖掘步骤。为此,提出一种基于GPS数据的驾驶行为异常检测方法。利用时间、速度、加速度、方向、转角等全局与局部特征及其对应的统计量,构建车辆驾驶行为的特征属性,并基于多特征对已有的商用车轨迹数据进行聚类分析,得到区域性车辆异常驾驶行为检测结果。实验结果表明,该方法能够准确判断待测车辆的超速、急加速/减速、频繁变道等典型异常驾驶行为。

关键词: 异常驾驶行为, 凝聚层次, 聚类, 多特征, 结构距离, 拉普拉斯变换

Abstract:

At present,the abnormal behavior detection based on video mainly focuses on the single vehicle restricted scene,which is difficult to monitor the whole transport process.Meanwhile,the track analysis of GPS is mainly based on the prior threshold judgment,which lacks the steps of in-depth data analysis and information mining.Aiming at these problems,this paper proposes an abnormal detection method of driving behavior based on GPS data.It uses the global and local features such as time,speed,acceleration,direction and rotation angle and their corresponding statistics,to construct the characteristic attributes of the driving behavior of the vehicle.It clusters and analyzes the existing commercial vehicle trajectory data to get the result of abnormal driving behavior detection,which is based on above features.Experimental results show that this method can accurately determine typical abnormal driving behaviors such as overspeed,rapid acceleration/deceleration and frequent lane change of the vehicle under test.

Key words: abnormal driving behavior, agglomerative hierarchy, clustering, multi-feature, structural distance, Laplacian transform

中图分类号: 

• TP391.7