基于光照不变性的车道线检测与跟踪算法研究

发布时间：2018-08-24 17:35

【摘要】：由于国内交通快速发展,引起的负面影响就是交通事故急剧增加,其中有许多由于车道偏离引起的交通事故,因此实时性高、可靠性强的车道线检测与跟踪成为了车辆导航性能要求的主要内容。近年来,由于许多研究人员的努力,在这一领域已经取得了一些进展,例如,应用在高速公路场景中的道路识别已经非常成熟。本文对车道线的检测与跟踪进行了研究,其中车道线检测被广泛应用于自动驾驶和防撞报警系统中。车道线检测系统即在道路图像中,通过预处理算法排除干扰,以及初步的对图像进行整理,提取出有效的车道线信息,且将其识别。本文车道线检测主要包括预处理、车道线检测算法、车道线跟踪算法以及CSK算法改进四部分。(1)车道线检测算法。通过预处理算法,经过逆透视变换,高斯滤波以及分位数方法,对于不同光照亮度的图像实施车道线检测做准备。然后对于不同光照亮度的图像做灰度化处理,车道线识别主要运用了改善的快速随机抽样的线性拟合一致性。(2)车道线跟踪算法。本文研究了卡尔曼滤波算法和CSK(Exploiting the Circulant Structure of Tracking-by-Detection with Kernels)跟踪算法,同时发现CSK算法在目标被遮挡时,无法实现跟踪,为此对CSK算法不防遮挡进行了改进研究。(3)算法测试。根据本文的算法,车道检测和跟踪测试的真实场景。检测结果表明,所提出的检测算法能够准确、快速的实现车道线检测;跟踪结果表明,卡尔曼滤波以及CSK跟踪算法对比分析实验数据,CSK比Kalman跟踪的效率高,速度快,而且改进后的CSK算法能够成功实现遮挡时目标的跟踪。
[Abstract]:Due to the rapid development of domestic traffic, the negative impact is the sharp increase of traffic accidents, many of which are caused by lane deviation, so the real-time performance is high. High reliability lane detection and tracking has become the main content of vehicle navigation performance requirements. In recent years, due to the efforts of many researchers, some progress has been made in this field, for example, the road recognition used in freeway scene is very mature. In this paper, the detection and tracking of lane lines are studied, in which lane detection is widely used in automatic driving and collision alarm systems. Lane detection system in the road image, through the pre-processing algorithm to eliminate interference, and preliminary collation of the image, extract effective lane information, and identify it. This paper mainly includes four parts: pretreatment, lane detection algorithm, lane tracking algorithm and improved CSK algorithm. (1) Lane line detection algorithm. Through pre-processing algorithm, inverse perspective transformation, Gao Si filter and quantile method, the lane detection of images with different illumination brightness is prepared. Then the grayscale image with different illumination brightness is processed, the lane line recognition mainly uses the improved linear fitting consistency of the fast random sampling. (2) the lane line tracking algorithm. In this paper, the Kalman filter algorithm and CSK (Exploiting the Circulant Structure of Tracking-by-Detection with Kernels) tracking algorithm are studied, and it is found that the CSK algorithm can not achieve tracking when the target is occluded. Therefore, the CSK algorithm is improved. (3) algorithm test. According to this algorithm, lane detection and tracking test of the real scene. The results show that the proposed algorithm can detect the lane accurately and quickly, and the tracking results show that the Kalman filter and the CSK tracking algorithm are more efficient and faster than the Kalman tracking algorithm in comparing and analyzing the experimental data. Moreover, the improved CSK algorithm can successfully achieve occlusion target tracking.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.6;TP391.41

【参考文献】