【作者】 张璐；

【导师】 刘纪红；

【作者基本信息】 东北大学 ， 电路与系统， 2014， 硕士

【摘要】 近年来,随着交通的发展和车辆的持续增多,交通安全事故频频发生,安全驾驶已成为日益突出的问题。智能交通系统(Intelligent Transportation Systems, ITS)与车辆辅助驾驶系统(Driver Assistance Systems, DAS)成为国内外研究的热点。其中,前方车辆的实时检测技术是车辆辅助驾驶系统中的关键内容之一,对于实现停车辅助、碰撞避免等具有十分重要的意义。本文设计了一种基于车牌的动静态车辆检测算法,开发了实时车辆距离预警系统。综合利用车辆的车牌特征、纹理特征、灰度对称性特征以及统计特征来检测前方车辆,并将该算法在DSP应用板中编程实现。在算法研究方面,本文利用实际生活中最为常见的蓝色和黄色车牌在YUV颜色空间的聚类性,实现了车牌提取,利用腐蚀、膨胀的开闭运算等形态学处理方法对车牌图像进行了降噪和修复,并结合连通域标记方法实现了对多个车牌的标记和定位,利用基于知识的纹理及灰度对称性特征完成对非车辆区域的排除。为解决车辆遮挡问题,本文设计了构造车辆局部特征子空间的方法,并采用二维主成分分析(Two-Dimensional Principal Pomponent Analysis,2DPCA)与最小距离分类器进行车辆的验证,实现车辆检测；最后,本文建立了车距预警模型,结合检测算法,实现了近距离车辆的警示,并对警示车辆的左、右侧方位进行判别。本文首先在MATLAB平台上完成了视频车辆检测算法的研究和仿真测试。之后,在硬件实现方面,分析了系统硬件需求,选取ICETEK-DM6437-B评估板作为硬件平台,选用CCD摄像头作为视频采集工具；其次,在车牌提取与初步定位阶段,将图像进行抽行抽列处理,减少了DSP处理的数据量,提高系统检测效率；最后,将算法移植到CCS集成开发环境中,利用C语言编程并进行了代码的优化,完成视频车辆的检测与车距预警。实验结果表明,本文设计的车辆检测与预警算法检测率可达90%以上,具有一定的鲁棒性和实用性。经过代码优化,效率得到了显著提升,每秒钟能完成大约三帧数据的处理,具有较好的实时性。并且该算法能有效的检测存在遮挡的车辆,减小光照等外界条件的影响。更多还原

【Abstract】 In recent years, with the development of the traffic and the increasing number of vehicles, traffic safety accidents increase frequently. Defensive driving has become an increasingly prominent problem. Intelligent Transportation Systems (ITS) and the Driver Assistance Systems (DAS) have become a hot research topic at home and abroad. Real-time detection technology of the vehicles at the front is one of critical content belonging to Driver Assistance Systems. It has very important significance to realize the parking assistance and collision avoidance.This thesis has designed a moving and stationary vehicles detection algorithm for real-time image and developed a real-time vehilces distance early warning system. The algorithm has utilized vehicles plate features, texture features, gray symmetry features and statistical characteristics. It is programmed on DSP in the end.In the aspect of algorithm research, firstly use the clustering of blue and yellow plate which is the most common in the actual life in YUV color space to extract the plate and process the binary image by corrosion expansion, connected component labeling, to realize the initial location of the vehicles area. Then use the symmetry texture features and gray preliminary to exclude non-vehicles. Besides, structuring local feature subspace of the vehicles, it can solve the problem of vehicle obscured effectively, and then use Two-Dimensional Principal Pomponent Analysis (2DPCA) and the minimum distance classifier vehicles verification to realize the final vehicles detection and localization. Finally, a vehicles distance early warning module is established and it is combined with the vehicles detection algorithm to alarm when the vehicles detected is very close, and determines vehicles warned left and right orientation.This thesis has completed video vehicles detection algorithm and simulation tests in the MATLAB platform firstly. In terms of hardware implementation, firstly, the thesis analyzes the hardware requirements, and selects ICETEK-DM6437-B evaluation as a hardware platform and selects the CCD camera as a video capture tool. Secondly, at the stage of extracting plate area and locating the vehicles, the thesis uses the method of down-sampling the image to reduce the amount of data processed by the DSP, and the efficiency of detection system is significantly improved. Finally, the algorithm is completed in CCS integrated development environment, using the C language to realize the vehicles detection and vehicles distance early warning real-time.Experimental results show that the video vehicles detection and warning system designed has an above 90% accuracy rate, with a considerable robustness and practicality. The efficiency of optimized code has been significantly improved to process almost three frames per second, which is capable to meet the basic real-time requirements. Besides, the algorithm can effectively detect the vehicles which are obscured partly, reducing the influence of light and other external conditions.更多还原