【作者】 刘波；

【导师】 聂明新；

【作者基本信息】 武汉理工大学 ， 通信与信息系统， 2007， 硕士

【摘要】 在信息化飞速发展的今天，交通信息的采集对于交通智能化的管理扮演着越来越重要的角色。交通信息采集系统的核心是交通流检测技术，而交通流检测的关键在于车辆检测、车速判断和车型识别，作为智能化交通的重要组成部分，它广泛应用于收费系统、交通数据统计等相关工作中，特别是在高速公路自动收费系统上，车型的自动识别更是占有极其重要的位置，因为它不仅是决定高速公路运营效率的主要因素，而且还是收费站决定收取费用的重要标准之一。由于目前传统上最主要的方法是在公路上埋设电缆线及感应线圈，通过摄像头装置，抓拍进入视线的车辆的照片进行车型识别，其它的如超声波检测方法、微波检测方法、红外线检测方法等在车型识别上都有不同程度的应用，虽然这些方法已经相当成熟，但由于这些方法不是对路段有破坏性，设备后期维护要求高，就是不适合沿道路大量铺设。而随着音频信号识别技术的发展，这些问题将被逐渐解决，声音音频信号识别技术包括语音识别，车辆声音识别，机械噪声识别，水声识别等等，虽然它设备简单，使用方便，但由于技术原因还未达到令人满意的程度，因此与其他方式结合使用能够取得更好的效果。本文以车辆行驶时产生的声音音频信号为基础，主要对车辆音频信号的特征进行分析，在此基础上提出了基于车辆音频信号对车型进行识别的方案的理论研究。本文首先查阅了国内外相关研究的进展，概述了车型识别技术的发展状况及车辆音频信号的产生机理，对车辆音频信号和语音信号做了比较，分析了利用语音信号处理的方法对车辆音频信号进行处理的可行性，对车辆音频信号的特征参数的提取做了研究。识别过程中先做了理论研究，选用隐马尔可夫模型作为识别系统的基础，阐述了HMM模型的基本原理，简述了该模型的三个问题和基本算法，提出了基于车辆音频信号进行车型识别的方案流程，阐述了线性预测倒谱系数和美尔倒谱系数的Matlab提取实现，在此基础上，对车型识别研究方案进行了理论上的探讨，而在最后的识别过程中是采用车辆音频信号的短时能量、过零率和基音周期等特征参数进行初步识别。由于本文的重点主要在于分析车辆音频信号的特征，是为将来的车型识别研究做铺垫，所以在车型识别方面，仅做了方案的理论研究。更多还原

【Abstract】 With the electronic information technology’s flying development, the collection of transportation information becomes more and more important to the management of the transportation intellectualization. The core of transportation information collecting system is transportation flow detection, the key is vehicle detection, velocity estimation and vehicle recognition. As an important part of intellectualized transportation, it applies widely in charge system, data statistics and so on, especially in highway automatic charge system, vehicle recognition play an extremely important role, because it is not only the main factor that determine the efficient of road transportation, but also the one of the important criterions for charging. In the present, the main method is to berry cable and induction coil under road, use camera to grasp the vehicle picture to carry on recognition. In addition, other methods also used in vehicle recognition, such as ultrasonic, microwave, infrared. Although these methods have matured sufficiently, they either are destructive to road, cost too much for later maintenance, or don’t suit for use on a large area. With the development of sound recognition, these problem will be solved gradually, sound recognition including speech recognition, vehicle acoustic recognition, mechanical noise recognition, water acoustic and so on. Although it has simple equipment and convenient, it doesn’t reach the mature stage, so if it uses with another method, it will be play an important role. This paper primarily carries on the analysis to the vehicle acoustic characteristic, then, researched on the vehicle recognition scheme by moving vehicle acoustic signal.In this paper, the development of vehicle recognition technology has been outlined by going through a lot of material in newspaper and references firstly, and then makes a contrast between vehicle acoustic and speech signal after analyzing their mechanism of the generation. The feasibility which process vehicle acoustic signal using speech signal processing method had been analyzed, and have a research on the extracting of vehicle’s characteristic. This paper chose hidden Markov model as the foundation of recognition system, its principle is presented, give a sketch of three problem and basic algorithm, put forward a scheme procedure for vehicle recognition based on vehicle acoustic signal, discussed the realization of LPCC and MFCC with Matlab speech toolbox in detail. But in the recognition, this paper uses short-time energy, zero-crossing rate and pitch period as the basic of recognition. Since the paper is focus on analyzing vehicle’s acoustic signal, it is the basic of recognition, so it only did a plan research in recognition.更多还原