基于图像处理技术和NBClassifier的车型识别技术研究-交通信息工程及控制专业论文.docxVIP

五邑大学硕士学位论文摘 五邑大学硕士学位论文 摘 要 智能交通系统(ITS)是现代交通发展的必然趋势，基于视频技术的车辆类型自动识别 是智能交通系统的一个重要研究方向，在城市道路交通监控和高速公路收费等系统中有 着广泛的应用前景。本文对智能交通系统中的车型识别进行了深入地研究，主要工作如’ 下： (1)研究视频图像预处理算法，对比分析研究几种常用的图像滤波方法，针对交 通视频环境的特点，用快速中值滤波法进行视频图像的噪声消除。 (2)针对交通图像的特点，给出一种行快速扫描算法，其基本思想是：在AOI区 域内，先通过逐行扫描获得最大的宽度确定为检测车辆的宽度。在已经确定的宽度区域 以确定车辆宽度的那行作为基准行，向两个方向再次进行逐行扫描，扫描结束的条件是 搜索到整行白点，记录行坐标，将两个行坐标的差确定为检测车辆的长度。实验证明该 算法在复杂的交通环境中表现优秀，有很好的实时性和准确性。 (3)将朴素贝叶斯分类器应用于车型分类，并且根据车型分类特点建立模型。将 各汽车制造商给出的车辆数据作为样本进行训练，使用本系统检测的车辆数据进行测 试，并与前向神经网络的测试效果进行比较。实验结果表明，朴素贝叶斯分类器在车型 分类方面具有较强的能力。 关键词：车型识别；车辆检测；朴素贝叶斯；图像处理； 五邑大学硕士学位论文Abst 五邑大学硕士学位论文 Abst ract Intelligent Transport Systems(ITS)is the determinate trend for the development of the modem transports，the vehicle automatic classification based video technology is important research field for the advance of the ITS．It has promising prospect in the application of the road tra伍c surveillance system and the highway toll system and on．The technique about vehicle automatic classification system is researched deeply．The main research work of this paper presented as follows： (i)Traffic video Image preprocessing．By analyzing several common images filtering，the preprocessing of traffic images by using median filtering are proposed． (2)This paper proposes new fast range scanning method to pick up the characteristic ofthe moving vehicle．The basic principle is that in the AOI all the lengths of all the detected targets gained through line scan and regard the largest width as the width of the vehicle．Scan in the gained width until all the pixels white line，then the length of the vehicle Can be obtained．The result shows that the algorithm also obtain precise characteristic values． (3)This paper puts NBClassifier theory into traffic vehicle automatic classification system．Use the actual vehicle characteristic data of length and width as training samples to train NBClassifier off online． The standard vehicle images captured from CCD camera,with the vehicle characteristic data of length and width obtai