基于SVM的地磁车辆检测器车型分类方法研究

发布时间：2018-06-23 09:25

本文选题：车型分类 + 地磁传感器　；参考：《北京交通大学》2014年硕士论文

【摘要】：随着城市化进程的推进,私人车辆日益普及,公路交通事业飞速发展,智能交通系统的研究受到非常大的重视。车型自动分类在智能交通系统中占有核心的地位,可以广泛应用于交通规划、路网设计和交通管理等相关工作中,具有广阔的应用场景。AMR地磁感应车辆检测器具有小体积、低成本、高灵敏度、安装维修方便等优点,是本文的研究对象。本文详细的研究了AMR地磁感应车辆检测器的原理,在前人依靠双节点进行车型分类的基础上进行了改进,在单个采集节点上使用了三轴的AMR传感器,传感器的方向正交并分别对应车高、车宽和车长三个方向,这种设计方式丰富了地磁采集信息,综合考虑了车辆构造和形状对磁场的扰动,减弱了车速对车型分类的影响,使得单节点实现车型分类成为可能。深入分析了AMR传感器采集到磁场强度信号的特点,在此基础上使用动态基准值的方法将车辆地磁扰动信号分离开来,提取了地磁信息的特征并使用Filter-Filter-Wrapper混合模型方法进行特征优化。对各种多分类SVM算法进行了比较和研究,选择有向无环图支持向量机作为本文的车型分类算法,并对传统DAG-SVM算法进行了改进,理论上证明了改进算法能够降低分类误差。为了验证本文所提出的车型分类算法的有效性,在增加视频采集的基础上搭建了一个车型分类验证系统,并在北京交通大学校内和校外道路上进行了实地实验。实验表明,本文设计的单节点地磁车型分类检测器的车型分类效果较好,具有较高的识别率,达到了预期目标。
[Abstract]:With the development of urbanization, private vehicles are becoming more and more popular, and highway traffic is developing rapidly. The research of Intelligent Transportation system (its) has been paid great attention to. Automatic vehicle classification plays a key role in intelligent transportation system. It can be widely used in traffic planning, road network design and traffic management, and has a broad application scenario. AMR geomagnetic induction vehicle detector has a small volume. The advantages of low cost, high sensitivity, convenient installation and maintenance are the object of this paper. In this paper, the principle of AMR geomagnetic induction vehicle detector is studied in detail, which is improved on the basis of previous vehicle classification based on two-node points, and a three-axis AMR sensor is used on a single acquisition node. The direction of the sensor is orthogonal and corresponds to the three directions of vehicle height, width and length respectively. This design method enriches geomagnetic information, synthetically considers the disturbance of vehicle structure and shape to the magnetic field, and weakens the influence of vehicle speed on vehicle classification. This makes it possible for single node to realize vehicle classification. The characteristics of magnetic field intensity signal collected by AMR sensor are analyzed in depth. On this basis, the vehicle geomagnetic disturbance signal is separated by the method of dynamic reference value. The features of geomagnetic information are extracted and optimized by using Filter-Filter-Wrapper hybrid model. This paper compares and studies various multi-classification SVM algorithms, selects directed acyclic graph support vector machine as the vehicle classification algorithm in this paper, and improves the traditional DAG-SVM algorithm, which theoretically proves that the improved algorithm can reduce the classification error. In order to verify the validity of the vehicle classification algorithm proposed in this paper, a vehicle classification verification system is built on the basis of adding video collection, and field experiments are carried out on the campus and off-campus roads of Beijing Jiaotong University. The experimental results show that the single-node geomagnetic vehicle classification detector designed in this paper has better vehicle classification effect, higher recognition rate and achieved the expected goal.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495

【参考文献】