车标自动精确定位算法研究
发布时间：2019-05-24 07:49
【摘要】：车辆识别技术是智能交通系统技术的重要关键技术之一,在军事车辆管理中有重要应用价值。车标定位识别技术是车辆识别技术的重要研究领域。车标定位技术往往需要车牌定位作为重要的辅助,快速、准确、高鲁棒性的车牌定位、车标定位是智能交通系统的研究热点。本文提出两个具有颜色不变性的颜色比例用于车牌颜色提取,实现车牌快速、准确、鲁棒性定位,完成车标粗定位;提出车标定位改进算法,提高定位精度。在基于车牌定位的车标粗定位中,基于RGB颜色模型本文提出了两个颜色比例,基于双色反射模型从理论上证明了它们具有颜色不变性,并首次采用这两个颜色比例提取车牌颜色特征,用于基于全局颜色特征和局部边缘特征相结合的车牌定位算法中。对1078幅交通摄像头实拍车辆图像进行处理,车牌定位成功率为99.9%。车辆原图大小约1000×1000像素时,车牌平均定位时间约0.2秒。与常用的基于HSV颜色模型的车牌定位方法相比,后者定位成功率是92.12%,平均定位时间约0.35秒。本文方法明显好于基于HSV模型的车牌定位方法。经比较发现本文颜色比例提取蓝色和HSV模型提取蓝色可以实现优势互补。利用车牌定位结果进而实现车标粗定位。快速、高效的车标识别需要准确的车标定位。为进一步抑制栅格背景干扰,提高车标定位精度,本文提出基于边缘梯度角度直方图分析和局部多阈值处理相结合的车标定位改进算法。分析边缘梯度角度直方图,将车标同栅格背景进一步分离。采用局部多阈值处理的方法消除光照不均的影响,得到更完整的车标。对5个品牌车辆67幅车标图像进行处理,所有车标均能较完整、准确地定位,车标定位率为100%。与常用的基于模板匹配的车标定位方法相比,后者虽能定位出所有车标,但误差较大的有28幅,占41.79%。本文方法定位精度明显好于模板法的有53幅,占79.1%。对模板法误差图像逐一分析发现模板法对栅格背景要求更高。以上结果说明本文方法定位效果优于基于模板匹配的车标定位方法,鲁棒性更强。车标粗定位区域约为260×170像素时,本文方法平均定位时间为0.1秒,模板法平均定位时间为0.04秒。模板法比本文方法要简单,速度更快。不过本文方法基本能满足实时性要求。
[Abstract]:Vehicle recognition technology is one of the important key technologies of intelligent transportation system technology, and it has important application value in military vehicle management. Vehicle mark location and recognition technology is an important research field of vehicle recognition technology. Vehicle mark location technology often needs license plate location as an important auxiliary, fast, accurate and highly robust license plate location. Vehicle mark location is the research focus of intelligent transportation system. In this paper, two color ratios with color invariance are proposed for license plate color extraction to realize fast, accurate and robust location of license plate, and to complete the rough location of vehicle mark, and an improved algorithm of vehicle mark location is proposed to improve the positioning accuracy. In the rough location of vehicle marks based on license plate location, based on RGB color model, two color ratios are proposed in this paper. Based on the two-color reflection model, it is proved theoretically that they have color invariance. For the first time, these two color ratios are used to extract license plate color features, which are used in license plate location algorithm based on the combination of global color features and local edge features. 1078 vehicle images taken by traffic camera are processed, and the success rate of license plate location is 99.9%. When the original size of the vehicle is about 1000 脳 1000 pixels, the average positioning time of the license plate is about 0.2 seconds. Compared with the common license plate location method based on HSV color model, the success rate of the latter is 92.12%, and the average positioning time is about 0.35 seconds. This method is obviously better than the license plate location method based on HSV model. It is found that blue extracted by color proportion and blue extracted by HSV model can complement each other. The rough location of vehicle mark is realized by using the result of license plate location. Rapid and efficient identification of vehicle signs requires accurate positioning of vehicle signs. In order to further suppress the grid background interference and improve the positioning accuracy of vehicle marks, an improved algorithm based on edge gradient angle histogram analysis and local multi-threshold processing is proposed in this paper. The edge gradient angle histogram is analyzed, and the vehicle mark is further separated from the grid background. The local multi-threshold processing method is used to eliminate the influence of uneven light, and a more complete vehicle mark is obtained. The images of 67 signs of 5 brand vehicles are processed, and all the signs can be located completely and accurately, and the positioning rate of the signs is 100%. Compared with the commonly used vehicle mark location method based on template matching, the latter can locate all the vehicle marks, but 28 of them have large errors, accounting for 41.79%. The positioning accuracy of this method is obviously better than that of template method, accounting for 79.1%. It is found that the template method requires higher grid background by analyzing the error images of template method one by one. The above results show that the localization effect of this method is better than that of the vehicle mark location method based on template matching, and the robustness of the proposed method is stronger than that of the template matching method. When the rough positioning area of the vehicle mark is about 260 脳 170 pixels, the average positioning time of this method is 0.1 seconds, and the average positioning time of the template method is 0.04 seconds. Template method is simpler and faster than this method. However, this method can basically meet the real-time requirements.
【学位授予单位】：北京理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495;TP391.41
本文编号：2484691
[Abstract]:Vehicle recognition technology is one of the important key technologies of intelligent transportation system technology, and it has important application value in military vehicle management. Vehicle mark location and recognition technology is an important research field of vehicle recognition technology. Vehicle mark location technology often needs license plate location as an important auxiliary, fast, accurate and highly robust license plate location. Vehicle mark location is the research focus of intelligent transportation system. In this paper, two color ratios with color invariance are proposed for license plate color extraction to realize fast, accurate and robust location of license plate, and to complete the rough location of vehicle mark, and an improved algorithm of vehicle mark location is proposed to improve the positioning accuracy. In the rough location of vehicle marks based on license plate location, based on RGB color model, two color ratios are proposed in this paper. Based on the two-color reflection model, it is proved theoretically that they have color invariance. For the first time, these two color ratios are used to extract license plate color features, which are used in license plate location algorithm based on the combination of global color features and local edge features. 1078 vehicle images taken by traffic camera are processed, and the success rate of license plate location is 99.9%. When the original size of the vehicle is about 1000 脳 1000 pixels, the average positioning time of the license plate is about 0.2 seconds. Compared with the common license plate location method based on HSV color model, the success rate of the latter is 92.12%, and the average positioning time is about 0.35 seconds. This method is obviously better than the license plate location method based on HSV model. It is found that blue extracted by color proportion and blue extracted by HSV model can complement each other. The rough location of vehicle mark is realized by using the result of license plate location. Rapid and efficient identification of vehicle signs requires accurate positioning of vehicle signs. In order to further suppress the grid background interference and improve the positioning accuracy of vehicle marks, an improved algorithm based on edge gradient angle histogram analysis and local multi-threshold processing is proposed in this paper. The edge gradient angle histogram is analyzed, and the vehicle mark is further separated from the grid background. The local multi-threshold processing method is used to eliminate the influence of uneven light, and a more complete vehicle mark is obtained. The images of 67 signs of 5 brand vehicles are processed, and all the signs can be located completely and accurately, and the positioning rate of the signs is 100%. Compared with the commonly used vehicle mark location method based on template matching, the latter can locate all the vehicle marks, but 28 of them have large errors, accounting for 41.79%. The positioning accuracy of this method is obviously better than that of template method, accounting for 79.1%. It is found that the template method requires higher grid background by analyzing the error images of template method one by one. The above results show that the localization effect of this method is better than that of the vehicle mark location method based on template matching, and the robustness of the proposed method is stronger than that of the template matching method. When the rough positioning area of the vehicle mark is about 260 脳 170 pixels, the average positioning time of this method is 0.1 seconds, and the average positioning time of the template method is 0.04 seconds. Template method is simpler and faster than this method. However, this method can basically meet the real-time requirements.
【学位授予单位】：北京理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495;TP391.41
【参考文献】
相关期刊论文 前4条
1 王泽兵,杨朝晖;彩色图像分割技术研究[J];电视技术;2005年04期
2 李贵俊,刘正熙,游志胜,王宁;基于能量增强和形态学滤波的车标定位方法[J];光电子·激光;2005年01期
3 李文举,梁德群,王新年,于东;基于纹理一致性测度的汽车车徽分割方法[J];计算机应用研究;2004年10期
4 李红林;王运琼;;基于差分与对称性检测相结合的车标定位方法[J];曲靖师范学院学报;2008年06期
，本文编号：2484691
本文链接：https://www.wllwen.com/kejilunwen/daoluqiaoliang/2484691.html
