汽车牌照，道路交通标志的自动分割和识别

【作者】 谢志鹏；

【导师】 陈锻生；

【作者基本信息】 华侨大学 ， 计算机应用， 2001， 硕士

【摘要】 汽车牌照，道路交通标志的自动分割和识别（Plate License＆＆Road SignAutomatic Extraction and Recognition）是智能交通系统（ITS）的重要软件支撑，具有广泛的应用前景。 文章分为两部分，第一部分为汽车牌照的自动分割和识别；第二部分为道路交通标志的自动分割和识别。 牌照部分的分割和识别策略：本文采用数学形态学运算，使用不同尺度和形状的结构元素来去除复杂背景的水平高频干扰分量，并使汽车牌照区域聚集成长条状，而后利用种子搜索区域测量法来提取牌照区域，接着用Hough直线检测来计算牌照的倾斜角度，在牌照倾斜矫正后，利用水平的高频分量分析和投影法对牌照的边框和铆钉进行去除，而后采用轮廓投影法分离出单个可供识别的车牌字符，在对单个车牌字符进行细化并膨胀后，获取其骨架特征点并组成一特征点向量；并对每个字符图象进行小波分解，以获得横向，纵向和斜向的笔画边沿图象，然后计算每个边沿图象的不变矩，获得三个不变矩向量；最后，将上述四个向量作为特征参数输入BP神经网络进行牌照号码训练学习和识别。 路标部分的分割和识别策略：本文首先采用基于HIS彩色模型的分割方法来提取出类似于路标颜色的物体，然后利用基于统计的灰度共生矩阵纹理分析，将表面纹理粗糙的物体过滤掉，留下表面纹理平滑的路标。然后利用傅立叶描述子提取出矩形，圆形，三角形的路标。而后对三角形，矩形和在直径方向上带长条形粗禁令标志的圆形路标，进行旋转矫正并规格化后，计算各子区域墒，组成一墒向量，而后对上述路标和直径方向上不带长条形粗禁令标志的圆形路标，计算其不变矩，组成一不变矩向量，利用上述的两个向量，在路标图案的特征库中利用最小欧氏距离法进行查找匹配。 本系统的特点是可对不同复杂背景下的牌照和路标进行较准确的提取，达到较快的的分割和识别要求。更多还原

【Abstract】 The Automatic Extraction and Recognition of Car License Plate Number andRoad Ttraffic Sign is the fundamental software support of Intelligence Traffic System0 It has a wide Application Perspective0This Paper is divided into two section,the first section of automatic Extraction and Recognition is on Car License Plate Number, the second is on the Road Traffic Sign.The Strategy of License Plate section : First step: using the mathematic morphorlogy Operation by the structural element with different size and shape, get rid of the horizontal interference high frequency component, make the car license Plate number aggregate together with a long bar shape; Second Step: Extract the plateArea by the seed searching && shape analyzing method;Third Step: use the Hough Line dedection to calculate the slant angle of plate ,after putting right the slant plate,Use the horizontal high frequency analysis and projection method to get rid of the Outer Bounding frame and rivet . Fourth Step: use the contour projection to separate the Single car license plate character for recognition0 Fifth Step: get the skeleton of character ,after dilating it ,acquire the feature point and compose a feature point vector: use the wavelet decomposition to get the horizontal,vertical and diagonal edge gradient picture ,and calculate the invariant moment of each picture,thus form three moment vector. The last step: input these four vector mentioned above to the BackPropogation Neural Network to train and recognize the character.The strategy of road sign section: First step: use the Color Model based on Hue Saturation Illuminance to extract the object ,which has similar color of RoadSign ,from the background picture; Second Step: use the the statistical GrayCo-occurrence Matrix to filter out the object with rough surface texture ,and keep the road sign obj ect with smooth surface texture . Third Step: use the Fourier shapedescriptor analysis to take out the rectangular, round and triangle road sign object0 Fourth Step: as to the slant rectangular ,triangle object and round object which has a thick banning bar across the center, Put Right them and calculate its unit entropy toform an entropy vector0 Fifth Step: calculate the invariant moments of each road sign object ,including the round object which hasn抰 a thick banning bar across thecenter ,and form a moments vector0 Use these two vectors to classify the road sign by exploring the feature library file,the classification method is based on the minimumdistance method.The characteristic of this system is being able to extract correctly the car licenseplate and road sign under different complex background~ and meet the requirement of segmentation and recogniton in relatively short time.更多还原