自然环境下道路交通标志的检测与识别

发布时间：2018-01-19 02:38

本文关键词： 交通标志检测与识别最大稳定极值区域感兴趣区域提取多特征融合 SVM分类　出处：《山东大学》2017年硕士论文　论文类型：学位论文

【摘要】：据相关机构统计全世界每年有130万人左右因为道路交通事故而丧失珍贵的生命,其中与驾驶员自身因素相关的酒后或疲劳驾驶、超速行驶等成为了这些交通安全事故的主要诱因。交通事故不仅会造成巨大的经济损失,更重要的是会无情地夺取人类宝贵的生命,因此道路交通安全问题已不再是某个国家面临的问题,而是需要全世界各国共同解决的。为了有效提高道路交通安全和运输效率,降低事故发生频率,保障人们的人身财产安全,智能交通系统应运而生。交通标志识别系统是智能交通系统诸多先进技术领域中的一个重要分支,在无人驾驶车辆、智能机器人、辅助驾驶系统、辅助道路标志规划、导盲机器人等方面都具有广阔的发展和应用前景。因此对于交通标志识别系统相关技术的研究和探索非常具有学术意义和实用价值。本文以城市道路中常见的指示、禁令以及警告标志为研究对象,针对大场景自然环境下的道路交通标志的检测与识别问题展开研究和讨论,主要从高分辨率大场景下的快速交通标志检测、多类别交通标志的鲁棒识别和交通标志识别系统平台的设计与搭建这三个方面作了深入研究和探索。在交通标志检测方面,为解决传统的基于机器学习的交通标志检测方法需要对每一个待检测子窗口进行处理而导致算法实时性欠佳的问题,提出了颜色增强下的MSER提取标志候选区域结合线性SVM的快速交通标志检测方法。该方法根据标志的颜色进行颜色增强,对增强图像提取MSER得到交通标志感兴趣区域,然后在大场景高分辨率图像的多尺度滑动遍历检测搜索过程中仅对包含交通标志候选区域的滑动窗口进行HOG特征的提取和SVM分类判别,而对非标志候选区域的滑动窗口则不进行特征提取和分类判别。实验结果表明:改进的MSER+HOG+SVM方法在获得了较高的检测准确率以及较低的误检率的前提下,运算速度上有较大提升,且鲁棒性较好。在多类别交通标志识别方面,提出了融合全局特征和局部特征的多特征交通标志分类识别方法,有效地提升了识别度。该方法首先分别提取能够描述标志图像内部纹理信息的LBP特征、表示标志图像形状信息的HOG特征以及描述图像粗略轮廓信息的全局Gist特征,然后采用线性组合方式,实现特征融合互补,并通过主成分分析方法进行数据降维,最后采用支持向量机分类器进行交通标志训练与识别。实验结果表明:相对于提取单一特征的交通标志识别方法,基于多特征融合的算法获得了更高的识别精确度,同时也满足实时性要求。最后,本文以轮式机器人为主要硬件基础,利用Microsoft Visual Studio 2010结合OpenCV开源视觉库设计了基于MFC对话框的交通标志识别系统应用程序以模拟行车驾驶环境。系统平台主要集成了图像采集与实时处理、标志检测、标志识别和机器人运动控制等功能模块。
[Abstract]:According to the statistics of relevant organizations, there are about 1.3 million people in the world who lose their precious lives because of road traffic accidents every year, including drunk or fatigue driving related to drivers' own factors. Speeding has become the main cause of these traffic safety accidents. Traffic accidents will not only cause huge economic losses, more importantly, will ruthlessly take away the precious lives of human beings. Therefore, the problem of road traffic safety is no longer a problem faced by a certain country, but needs to be solved by all countries all over the world. In order to effectively improve road traffic safety and transport efficiency, reduce the frequency of accidents. The intelligent transportation system (its) emerges as the times require. Traffic sign recognition system is an important branch in many advanced fields of intelligent transportation system, which is used in driverless vehicles and intelligent robots. Auxiliary driving system, auxiliary road sign planning. Blind robot has a broad prospect of development and application. Therefore, the research and exploration of traffic sign recognition system is of great academic significance and practical value. Show. Ban and warning signs as the research object, the detection and recognition of road traffic signs under the large scene environment is studied and discussed, mainly from the high resolution of the rapid traffic signs detection. The design and construction of robust recognition and traffic sign recognition system platform for multi-class traffic signs are studied and explored in detail. In order to solve the problem that the traditional traffic sign detection method based on machine learning needs to deal with every sub-window to be detected, which leads to poor real-time algorithm. A fast traffic sign detection method based on MSER and linear SVM is proposed, which is based on the color of the sign. The area of interest is obtained by extracting MSER from enhanced image. Then in the search process of multi-scale sliding traversal detection of large scene high-resolution images, only the HOG feature extraction and SVM classification are carried out on the sliding window containing traffic sign candidate area. But the sliding window of unmarked candidate region is not extracted and classified. The experimental results show that the improved MSER HOG is improved. The SVM method can obtain higher detection accuracy and lower false detection rate. In the aspect of multi-class traffic sign recognition, a multi-feature traffic sign classification and recognition method combining global features and local features is proposed. The recognition degree is improved effectively. Firstly, the LBP features which can describe the internal texture information of the logo image are extracted separately. The HOG features representing the shape information of the logo image and the global Gist features describing the rough contour information of the image are presented. Then the features are fused and complemented by linear combination. Finally, the support vector machine classifier is used to train and recognize traffic signs. The experimental results show that the traffic sign recognition method is relative to extracting a single feature. The algorithm based on multi-feature fusion achieves higher recognition accuracy and meets the real-time requirements. Finally, this paper takes wheeled robot as the main hardware base. Using Microsoft Visual Studio. In order to simulate the driving environment, the application program of traffic sign recognition system based on MFC dialog box is designed based on OpenCV open source visual library in 2010. The system platform mainly integrates image acquisition and real-time processing. Sign detection, sign recognition and robot motion control and other functional modules.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.6;U495;TP391.41

【参考文献】