【作者】 吴骏；

【导师】 杨兆选；

【作者基本信息】 天津大学 ， 信号与信息处理， 2007， 博士

【摘要】 智能交通系统(Intelligent Transportation System,简称ITS)是指将先进的信息技术、数据通讯传输技术、电子传感技术及计算机技术等有效地集成运用于整个地面交通管理系统而建立的一种在大范围内实时、准确、高效的综合交通运输管理系统。智能交通系统已成为21世纪现代化交通运输体系的重要发展方向。论文的主要工作是研究智能交通系统中的一些信息处理关键技术,包括:车牌字符识别算法、车体定位算法、检测线式视频车辆检测算法、视频交通检测系统设计方法。论文的创新点主要体现在以下四个方面:1)提出了基于脉冲耦合神经网络(PCNN)的车牌字符识别算法,首次将PCNN应用于车牌字符识别算法。以简化PCNN模型为基础,提取三种不同的图像特征,根据它们对数字、字母和汉字字符图像进行识别。实验结果表明,与传统的BP神经网络算法相比,此算法总体识别率更高,容错性更好,鲁棒性和稳定性更强,使用更加方便灵活,并能满足实际应用中对识别速度的要求。2)提出了基于相位信息的边缘投影车体定位算法。将图像的HSV色彩空间的信息与RGB色彩空间的信息结合,利用log Gabor小波滤波器组检测到的相位一致性进行车辆边缘检测,利用投影算法和车体区域判别算法对车体进行定位。与常用的车体定位算法相比,此算法受图像亮度、对比度、噪声和车辆阴影的影响小,而且适用于包含多部车辆的图像,定位准确率高。3)提出了改进型检测线式视频车辆检测算法。此算法分别利用图像的亮度和色度信息进行两级检测,能够有效提高原算法的车辆检测准确率,降低拒真率和认假率,而且满足实时性需求。4)提出了基于ARM与DSP双核体系结构的嵌入式视频交通检测系统设计方案。设计了两个系统单元间通信接口的硬件连接方案和驱动程序,并且介绍了系统的操作系统和应用程序的设计方法。此系统将32位嵌入式微处理器ARM与数字信号处理器DSP相结合,充分体现了双微处理器系统的优势。与常用的视频交通检测系统相比,此系统具有体积小、功耗低、成本低、稳定性好、操作方便、界面友好等优点,而且满足实时性要求。更多还原

【Abstract】 Intelligent Transportation System or ITS, is an transportation management system which integrates advanced information technology, data communication technology, electronic sensing technology, computer technology and all kinds of other advanced technologies synthetically and applies them to the whole ground traffic management system. ITS constructs an effective in large-scale, real time, exact, and efficient transportation system, and it has become an important development aspect of the modernized transportation system in the 21st century. The main work of the thesis is research on some key technologies of information processing in ITS, which includes: vehicle license character recognition algorithm, vehicle localization algorithm, virtual-line based video vehicle detection algorithm and video transportation detection system design method.The innovations of the thesis are embodied in four aspects as follows:1) A vehicle license character recognition algorithm based on pulse coupled neural network (PCNN) is proposed. It firstly applies PCNN in vehicle license character recognition. Based on simplified PCNN model, it extracts three different image features and utilizes them to recognize number, letter and Chinese characters. Compared with common algorithms based on BP neural network,this algorithm has advantages of higher total recognition rate, better fault tolerability and stability, stronger robustness, and is more convenient and flexible to use. Its recognition speed is fast enough to satisfy the demand on speed in applications.2) A contour projection vehicle localization algorithm based on phase information is proposed. It combines the information of the image in HSV color space and RGB color space, detects contour of vehicles based on phase congruency utilizing log Gabor wavelet filters, then uses projection method and vehicle area distinguishing algorithm to localize vehicles. Compared with common algorithms, it suffers smaller from the luminance, contrast of image, noise and shadows, and has higher accuracy. It is also suitable for images including more than one vehicle. 3) An improved virtual-line based video vehicle detection algorithm is proposed. It introduces two-level detection, utilizing luminance information and the chrominance information respectively. The improved algorithm can effectively increase the accuracy of old one and reduce the FRR and FAR. It can satisfy the request of real-time performance.4) A design of dual-cored embedded video vehicular detection system based on ARM processor and DSP is proposed. The method of hardware connection between two units and the driver program of the interface are introduced. The design methods of operating system and application programs are also presented. This system combines 32 bit embedded microprocessor ARM and digital signal processor DSP, and it sufficiently embodies the superiority of dual microprocessor system. Compared with common video vehicular detection systems,it has advantages of small volume, low cost, low power consumption, good stability and expandability, facility of operation and friendly interface. It can satisfy the demand on real-time.更多还原