【作者】 李文举；

【导师】 梁德群；

【作者基本信息】 大连海事大学 ， 通信与信息系统， 2005， 博士

【摘要】 图像处理技术在智能交通中应用的研究,是智能交通系统的重要前沿研究领域,具有十分重要的理论意义和应用价值。 图像处理技术在智能交通中的应用领域非常广阔,大体上可分为基于视觉的智能车辆导航、基于视觉的交通监控和基于视觉的交通管理三大应用领域。本文主要研究了后两大领域中的若干问题,主要包括彩色边缘检测、车牌定位、车牌字符分割、车徽分割和运动车辆的阴影检测。 彩色边缘检测在面向智能交通的图像处理技术中发挥着重要作用。目前主要存在两个问题,一是大多数算法在RGB彩色空间中实现,不能提供符合人类视觉颜色理解的边缘信息。含有色调、饱和度和亮度的颜色模型虽然更符合人对颜色的感知,但基于此颜色模型的彩色边缘检测的研究还较少,且已有算法对噪声敏感,抗噪性能差。因此,在色调、饱和度、亮度彩色空间内寻找对噪声具有鲁棒性的彩色边缘检测算法是要解决的一个问题;二是目前的彩色边缘检测算法没有考虑目标的先验知识,所求得的边缘数量较多,目标特征不易提取。因此,寻找基于目标特征的彩色边缘检测算法是要解决的又一个问题。 基于图像处理的车辆识别技术是智能交通领域的重要研究方向之一,目前的方法是基于车牌识别和车型识别,它们都存在算法的可靠性问题,而且车牌还存在易更换问题。要想可靠地识别车辆,必须最大限度地利用车辆提供的信息。因此,除了基于车牌和车型识别的车辆识别技术,是否还能找到其它的车辆识别方法是我们要解决的一个问题。车牌定位和字符分割是车牌识别系统的关键环节,但目前还存在着有待解决的难题。已有的车牌定位方法,当车牌底色与其周围颜色近似、车牌底色褪色或图像中的区域具有与车牌相似的几何和纹理特征时,有效定位率下降;而已有的字符分割方法在车牌图像质量退化时分割效果很不理想。因此,寻找更可靠的车牌定位方法与车牌字符分割方法是要解决的两个问题。 在基于视觉的交通监控中,运动车辆检测是要解决的首要问题,但阴影的存在常常使检测出现错误。因此,进行阴影检测是非常重要的。目前存在的问题是,阴影检测主要在RGB和HSV彩色空间中进行,在RGB空间中,R、G、B颜色分量更多还原

【Abstract】 Image processing technology plays today an ever-increasing role in ITS(Intelligent Transportation Systems).Applying image processing technology to ITS is a challenging field ,which has great theoretical significance and practical value.In ITS, image processing technology is applied to a wide variety of areas such as vision-based intelligent vehicle guidance, vision-based traffic surveillance and vision-based traffic management. In this thesis ,we focus on the latter two fields which mainly include color edge detection, vehicle license plate location ,license plate character segmentation, car emblem segmentation and shadow detection of moving vehicles.Color edge detection plays an important role in image processing technology for ITS . There are two problems now. One is most edge detection algorithms are based on RGB color space, so they can’t provide edge information that corresponds to human visual perception of color. Though the color model containing the color components of hue,saturation and luminance corresponds more closely to human perception of color, few approaches to edge edge detection based on this color model have been proposed in the literature. Moreover, the existing approaches are sensitive to noise. So a problem to be solved is to find color edge detection algorithms that are robust to noise in the hue, saturation and luminance.The other problem is that prior knowledge about objects is not considered in the existing edge detection algorithms. As a result of that, the amount of edge point detected is high and it is difficult to obtain the characteristics of the object. So another problem to be solved is to find a color edge detection algorithm based on the characteristics of the objects.Image processing-based vehicle recognition is one of the important research fields in ITS. The existing methods are all based on license plate recognition and car shape recognition. Their common problem is algorithm stability. And the license plates are easy to be changed. All information about vehicles should be used to recognize them reliably .A problem to be solved is to find a method to recognize vehicles besides license plate recognition and vehicle model recognition. Vehiclelicense plate location and character segmentation are critical steps in the license plate recognition system ,and yet there are difficult problems to be solved. The existing vehicle license plate location methods are not robust under the conditions that the background colors of the plates are similar to the surrounding colors, the background colors of the plates degrade and there is any region with similar geometry and texture features to plates in the images. The existing character segmentation methods are not robust to the poor vehicle license plate images. It is necessary to find reliable license plate location and character segmentation methods.In the vision-based traffic surveillance systems, moving vehicle detection is the principal problem to be solved, but the existing shadow reduces the detection accuracy. So detecting shadow is very important.The existing shadow detection methods are mainly based on RGB or HSV color space. In the RGB color space, red,green and blue components are generally correlated, so the image analysis is not satisfactory in each color component separately. In HSV color space, hue is meaningless when the intensity is very low or very high, and hue is unstable when the saturation is very low. So how to detect shadow is a problem to be solved.In this thesis ,we perform further researches on the problems above mentioned. The main contributions of the thesis are as follows:A new color edge detection approach based on direction region distance measure is proposed. The method uses hue orthogonalization to solve the problem of modulus 27i.Based on the order, orientation and structure of color distribution in the neighbourhood of an edge and the effect of a filter scale on edge detection, a direction region distance measure is defined for color edge detection. The direction region distance measures for various purposes can be obtained by different weights . Experimental results have shown it is effective to suppress noise.The new concept of the edge-color pair is proposed in this thesis. Color is a very important feature of the object. In many applications, there is a constant color pattern matching between background and the object itself. If the color features of the background and the object combine with the edges of the target, the object area can be highlighted and the non-object areas can be suppressed and it is very useful to objectsegmentation. In this thesis, according to constant color matching between background and object, the new concept of the edge-color pair is proposed. A new color edge detection method based on edge-color pairs is proposed and used to vehicle plate location. Experiments have shown it is verye ffective.A new approach for license plate location based on edge-color pair is proposed. Considering the given color match between background and character in a license plate, edges are detected by color edge detection method based on edge-color pair in a vehicle image so as to highlight the plate area and suppress the non-plate areas to the extreme in the first step of plate location ,and then the plate is extracted based on structural and texture features. The method can not only provide the color information of a plate but also extract the plate exactly under the conditions that background colors of the plates are similar to the surrounding colors , the background colors degrade and there is any region with similar geometry and texture features to the plates in the images.A new active vision-based license plate character segmentation approach is proposed. This approach improves the character segmentation reliability by applying both coarse to fine and closed-loop feedback techniques to the processing of character segmentation, and the non-uniform illumination correction and contrast enhancement operations are also introduced to minimize the effects of non-uniform illumination and poor contrast . Two-degree feedback is adopted in the processing of character segmentation. First, the scale-adaptive cubic B-spline wavelet transform is applied to outer -contour vertical distance of the character regions for coarse character segmentation ; second, the template matching character recognition feedback based on object occupancy rate is used for fine character segmentation .The experimental results show that the proposed algorithm is robust to character segmentation when dealing with non-uniform illuminated , contrast-poor .declining , dirt , character-touching and character-broken number plates.A new idea about vehicle recognition based on car emblem recognition is proposed. The car emblem is a label of a car. It contains important information of a vehicle not only its model but also its manufacturer. Moreover, a car emblem isdifficult to be changed. Integrating the information of the car emblem with the plate and the shape of a vehicle will increase the robustness of the vehicle recognition .Car emblem recognition is a new domain of vehicle recognition technology. And it is very important complement and development of the existing vehicle recognition technology based on license plate and car shape recognition.A car emblem segmentation method based on texture homogeneity measure is proposed. Considering the texture in left side of a car emblem is same as in right side, the texture homogeneity measure is defined based on wavelet transform. To a car emblem between the headlights of a car, its coarse segmentation is first performed by the position information between the car emblem and the plate, and then the texture homogeneity measure and the morphological filter are used to segment the car emblem finely.A new shadow suppressing method is proposed in moving vehicle detection. This method first uses inter-frame difference to remove the static backgrounds and obtains the vehicle contour and moving shadow contour, and then removes the correlation between R,G,B components by Hotelling transform, finally defines a shadow measure to detect the shadows and extract the vehicle contour.The above contributions constitute the main parts of this thesis. The content of this thesis includes the main aspects of image processing technology in ITS. Among these, the proposed color edge detection method is not only effective in license plate location, but also useful in color edge detection to given objects and provides a new idea for color edge detection. The other contributions are further development of image processing technology in ITS. The proposed new idea and methods have great guiding significance to the application research of image processing technology in ITS and have greater practical value.更多还原