【作者】 圣华；

【导师】 李钢；

【作者基本信息】 合肥工业大学 ， 计算机应用技术， 2010， 硕士

【摘要】 注意力分散是导致车道偏离而发生交通事故的常见原因之一。车道偏离预警系统实时探测车辆所在车道位置,若车辆偏离车道却没有开启相应的指示灯,系统将以光、声、振动等形式警告驾驶员,以保证行驶安全。由于路况的复杂性以及车辆行驶过程中要求系统具有较高的实时性,如何调节系统的鲁棒性、实时性和实用性以利于车道偏离预警系统能够投入到实际应用成为相关研究的重点。在汽车辅助安全驾驶系统中,准确获取道路信息,即车道标志线的轨迹,是实现准确无误报警的前提和基础。本文对车道线检测技术的基本概念和基本理论进行分析和论述,在分析研究了国内外已有的车道线检测算法的基础上,对一些经典的车道线检测算法进行分析和研究,并根据LMedSquare曲线拟合具有减少噪声的特点,设计一种LMedSquare选取最佳车道线种子点集的直线拟合算法,并讨论了在不同道路路况时的解决方案。在道路路况较好的情况下,待拟合的种子点数目比较充足时,提出利用本文设计的一种选取最优待拟合种子点的最佳子集选择方案；在道路有遮挡、油污污染而导致道路车道线信息量少的路况下,根据图像帧之间具有连续性、相关性的特点,充分利用此前多帧车道线参数对当前图像的车道线参数进行卡尔曼预测,把符合风险函数取得最小值的参数最为作为当前车道线的参数。该算法在车道线检测过程中,有效解决了道路车道标识线遭到干扰而导致无法识别的问题。本文研究、测试结果表明,基于LMedSquare选取最佳子集的车道线检测算法具有良好的检测效果,可应用到嵌入式集成视觉辅助行车安全系统实际项目中,实现辅助行车安全系统中的车道线参数的实时准确提取。更多还原

【Abstract】 The Distraction of your mind which led to lane departure is one of the common causes of traffic accidents. Lane departure warning systems can detect the location of vehicles Real-timely, where if the vehicle deviated from the lane and does not open the corresponding indicator, the system will warning drivers by light, sound, vibration and other forms of warning. As the complexity of road conditions and the process of car running demands the system has high real-time, how to adjust the system robustness, and practicality in order to put lane departure warning system into practical application to be the focus of the research.In the car safe driving system, accurate to obtain road information which is the path marking of the lane is the premise and foundation of accurate warning. In this thesis, the basic concepts and theories of lane detection technology was discussed.And analyzed some classics lane detection algorithm which has been studied abroad and home, and According to noise reduction of LMedSquare curve fitting, proposed that LMedSquare select the best set of seed points, and discussed the solutions of lane line detection algorithm in different road condition. In the favorable road condition, which caused the seed points are sufficient, design a options by which selected seed point of the best subset; when lane information is not clear which caused by barrier, oil pollution, in accordance with continuity and relativity successive the between image frames, we can full use of multi-frame line parameters which has been detected to predicted current lane parameters using Kalman filter, and than find the parameter which caused the risk function obtain the minimum value as the current lane parameters. This lane detection algorithm can detect lane parameter effective when lane has been disturbed.The research and test shows that the testing lane algorithm based on LMedSquare select the best set of seed points has good effect, and can be applied to the actual project of embedded system visual display traffic safety system, to distill the lane parameters of traffic safety system accurately in real time.更多还原