【作者】 林广宇；

【导师】 魏朗；

【作者基本信息】 长安大学 ， 载运工具运用工程， 2009， 博士

【摘要】 汽车保有量增加,公路交通拥挤,交通事故频发,公路交通安全问题日益突出,促使汽车安全辅助驾驶技术成为研究热点。利用传感器监控行车环境对处于危险状态的汽车及时报警的防撞安全预警系统,是提高道路交通安全的有效手段,是智能交通研究领域的一个重要组成部分。目前绝大部分安全辅助驾驶技术利用通用PC机主频高处理速度快的特点,采用PC机作为处理平台,然而,通用PC机冗余的功能、高额的成本和较大的体积是应用在车载系统中难以克服的瓶颈。嵌入式软硬件系统凭借实时性、低资源占用性以及体积小巧等特点,为汽车安全辅助驾驶系统的应用研究与开发提供了实用的技术手段。本文利用嵌入式技术对车道偏离预警以及最小安全车距预警进行了深入研究,有利于防撞安全预警系统的实用化和产品化,对于提高车辆的主动安全性,降低交通事故发生率,减小交通事故带来的危害具有重要意义。本文提出了利用驾驶员驾驶过程的数据分析驾驶员的驾驶特性,对于更加科学地分析驾驶员驾驶特性、加强驾驶员的培训与管理具有一定的理论意义和工程应用价值。论文以视觉传感器获取车辆前方环境信息,对道路标识线的识别与车辆偏离车道预警算法进行了研究,以检测前方车辆的方位为基础,建立最小安全车距预警模型预防汽车纵向碰撞的发生,根据车辆在行驶过程中的横向位置信息对驾驶员驾驶特性进行了分析与评价。最后,基于DSP和ARM芯片的双核硬件平台和Linux操作系统完成了嵌入式车载图像监控系统的设计与实现。利用计算机视觉理论和技术,研究了一种道路标识线的图像检测方法与车辆偏离车道预警规则。以经中值滤波、Sobel算子边缘检测和最大类间方差阈值分割预处理后的道路图像为基础,通过直线的矢量基元表示法对标准Hough变换加以改进实现道路标识线的提取与跟踪,提高了道路标识线的识别速度。基于摄像机内外参数的标定,依据车辆相对道路标识线的横向位置和横向偏转角提出了基于横向距离和横向速度的车道偏离预警算法,实验表明该方法能够充分考虑车辆的横向速度与动态偏转角度对车辆偏离的影响。基于单目视觉和透视投影变换原理,对防控车辆纵向碰撞的最小安全车距预警算法进行了研究。采用车辆阴影特征和车辆对称性特征的有条件联合方法识别检测前方车辆目标,利用Kalman滤波原理预测车辆在序列图像帧中的运动方位,从而实现对车辆的跟踪,缓和了车辆检测准确性要求与实时性要求之间的矛盾。以已识别车辆的矩形底边界为基础给出在结构化道路上车辆间纵向距离的计算方法；提出了基于前车状态的最小安全车距预警模型,实验证明,该方法缓解了理论计算的安全距离与驾驶员认知的习惯安全距离不相一致的矛盾。根据已检测出的道路标线以及车辆相对道路标线的横向距离,利用数据库知识发现理论(KDD)研究驾驶员在车辆行驶过程体现出的驾驶特性。建立驾驶员车道内行驶安全性评价模型,对驾驶员车道内行驶的横向距离数据进行统计分析,以模糊隶属度评价驾驶员车道内行驶的安全性。建立驾驶员换道超车行驶轨迹曲线模型并根据采样数据进行曲线拟合,利用信息融合技术对相似行驶轨迹曲线进行融合使得算法具有自适应性,提出换道超车操作稳定性评价模型,计算得出驾驶员换道超车稳定性的模糊隶属度。以真实交通环境为背景分析驾驶员的驾驶特性,克服了采用问卷调查方法和仪器测量方法分析驾驶员驾驶行为的主观性和局限性。最后,讨论了嵌入式车载图像监控系统的软硬件设计与实现,在基于数字信号处理器DSP和微处理器ARM9双核的评估板硬件平台上,应用Linux嵌入式操作系统对车载图像监控系统的总体功能进行了设计,完成相关功能模块的软件设计。更多还原

【Abstract】 With the increase of vehicle population, heavy traffic and too many traffic accidents make safety problems become increasingly prominent. Collision Avoidance Warning System, one of important components of Safety Driving Assist Systems, which alerts the vehicle in danger by using sensors to monitor driving environments, is one of effective measures to improve traffic safety. At present, most of Safety Driving Assist Systems have been studied using PCs which have high main frequency and processing speed as processing platforms. But redundant function, high cost and large size of PCs have restricted their application in on-board systems. Embedded systems for software and hardware provide practical technical means to study and develop Safety Driving Assist Systems, with their characters such as real-time performance, low resources use and small sizeResearch on Lane Departure Warning System and Minimum Safe Forward Distance Warning System by use of embedded techniques, is favorable for practicability and production of Collision Avoidance Warning System, which is of significance to improve active safety of vehicles, reduce traffic accidents and minimize losses brought by traffic accidents. This dissertation has also put forward methods of analyzing drivers’driving characters with data acquired in driving, which has important theoretical significance and engineering practical value to study more scientifically driving characters and strengthen training and supervision of drivers.This dissertation has discussed the recognition for road traffic marking lines and algorithm for lane departure warning, according to forward environment information obtained by a visual sensor. On basis of the location of forward vehicle, Minimum Safe Forward Distance Warning System is proposed to avoid automobile forward collision. A method of analysis and evaluation of Drivers’ driving characters is provided in accordance of lateral distance of the vehicle in the road. Finally, Embedded On-board Image Monitoring System is designed and realized on the hardware platform based on DSP and ARM chips and Linux operating system.Utilizing computer vision theory and technique, an image method of detecting road traffic marking lines and rules of lane departure warning are studied. Road images are preprocessed by median filter, Sobel operator edge detection and threshold segmentation of maximum class variances. Standard Hough Transform is improved by Elementary Line Segments representation in order to identify and track road traffic marking lines extraction more quickly. Based on camera calibration of interior and exterior parameters, lane departure warning algorithm is put forward dependent on lateral distance and lateral deflection angle of the vehicle to the road traffic marking line. The experiments results show that the algorithm can consider the effects of the vehicle’s lateral velocity and dynamic deflection angle to vehicle departure very well.By using monocular vision and projection transformation principle, minimum safe forward distance warning algorithm is proposed to avoid vehicle forward collision. Forward vehicle can be detected by means of conditional combination of shadow underneath a vehicle and grayscale symmetry of a vehicle. By use of Kalman Filtering principle, a vehicle’s moving positions in serial images are forecasted to track it, which has appeased the conflict of accuracy and real-time performance of vehicle detection. On basis of recognized vehicle rectangle hemline, a method of calculating longitudinal distance between vehicles on structured roads is put forward. Minimum safe forward distance warning model is established according to braking distance calculation in the automobile theory. The experiment results prove that the method ease up the conflict between safe distance from theoretical calculation and habitual safe distance of drivers.By use of Knowledge Discovery in Database, research on driving characteristics which a driver shows in traveling are discussed according as detected road marking lines and lateral distance of a vehicle to road marking lines. Safety evaluation model for driving in lane is built to analyze statistically lateral distance data in lane and safety driving in lane is evaluated by fuzzy membership grades. Driving routes model for changing lane to overtake is established and driving curves are fitted in accordance of sample data. Information fusion technique is used to fuse similar driving curves so as that the algorithm has self-adaptability. Stability evaluation model for changing lane to overtake is brought forward to account out fuzzy membership grades for changing lane to overtake. Analysis of driving characteristics in the background of real traffic environment overcome subjectivity of questionnaire survey and limitation of measuring with instruments.Finally, software and hardware for Embedded On-board Image Monitoring System are designed and realized. On hardware platform of evaluation board based on Digital Signal Processing (DSP) and microprocessor ARM9 and in Linux embedded operating system, overall function design of On-board Image Monitoring System is carried out, and software design of some important function modules is also realized.更多还原