摘要
机动车车型识别是城市道路交通流监测统计的一个重要方面。本文基于频谱分析与支持向量机方法提出一种车型音频识别方法,以1/3倍频程频谱数据作为特征数据,并使用支持向量机方法完成不同车型分类下的车型识别,同时还分析比较了不同训练样本量及不同单个样本数据量大小对识别结果的影响。在将车型细分的情况下,对小汽车、大型公交车、水泥车、摩托车四种车型的样本外识别结果达到96.9%的准确率,验证了方法的有效性。
Identification of vehicle type is the important content of the monitoring and statistics of urban road traffic flow.This article presents an audio identification method of vehicle type based on frequency analysis and support vector machine（SVM）.Researches began by exacting features as 1/3 octave spectrum data,which were then trained and classified by utilizing SVM classifier.Thereafter accuracy rates under various parameters such as number of training samples and size of single sample was analyzed and compared.An Out-of-Sample accuracy rate of 96.9% was achieved among 4 different types,which were car,heavy bus,cement truck and motorcvcle.
出处
《应用声学》
CSCD
北大核心
2014年第4期371-376,共6页
Journal of Applied Acoustics
基金
国家自然科学基金(51178476)
国家863计划课题(2012AA063301)
中国民航信息技术科研基地开放课题基金项目(201208)资助
关键词
车型识别
频谱分析
支持向量机
音频信号
Vehicle type identification
Frequency analysis
Support vector machine
Audio signal
