基于动态时间弯曲的多模板匹配车型分类

Multiple Templates Matching for Vehicle Classification Based on Dynamic Time Warping Algorithm

针对环形感应线圈采集数据的特点，讨论一种基于动态时间弯曲（DTW）算法的车型分类新方法。由于采样波形与车速、车辆自身结构以及其通过线圈时的相对位置有关，导致实际同类车辆波形存在伸长、压缩，甚至波形残缺不全或局部振荡等复杂情况。为此，本方法首先对样本数据进行降噪滤波，补齐不完整信号等预处理；然后结合DTW方法和最优聚类分析原理选取同类车型的多个模板，综合考虑误纳与误拒情况，可降低单模板匹配存在的误拒率；最后利用DTW算法和多个分类区分度指标进行多模板匹配，得出分类结论。论文仿真部分以一类车型为例，分析了本文方法与其他方法比较存在的优势。

According to the data properties of the inductive loop detectors, a novel method for the vehicle classification based on the detectors is discussed. For the sample wave of a vehicle is determined by its velocity, structural framework and the relative location to an inductive loop detector, the sample wave is flexible. To classify the vehicles, the pretreatment to the sample wave is necessary. A method of the multiple templates based on DTW and clustering analysis algorithms is used to improve the precision, and the classification result is educed by matching the sample wave to the multiple templates using a newly DTW algorithm. An experiment is given to analyze the effect of the method.