一种实时车道线偏离预警系统算法设计和实现

发布时间：2018-10-19 07:43

【摘要】：针对实时车道线偏离预警问题,采用一种横向腐蚀算子对边缘检测后的图像进行腐蚀,减少和消除图像中无关的边缘信息,从而显著减少后续处理数据量;提出一种以大津法为基础的边缘梯度图像分块阈值选取方法以便在不均匀光照条件下对道路边缘图像进行有效分割;结合车道线在路面分布的几何特征、Hough投票结果、道路图像之间的相关性和车道线宽度特征,提出了候选车道线筛选和计分算法对多车道场景进行车道线识别,采用卡尔曼滤波法对车道线进行跟踪,应用车道线偏离预警系统算法软件进行了试验验证。试验结果表明:道路图像总帧数为24 661,其中确检帧数为23 483,误检帧数为1 178,平均检测正确率为95.22%,因此,算法是正确的和有效的,可以较好地满足车道线偏离预警系统实时性和鲁棒性的要求。
[Abstract]:Aiming at the problem of real-time lane deviation warning, a kind of lateral corrosion operator is used to corrode the image after edge detection to reduce and eliminate the irrelevant edge information in the image, thus significantly reducing the data amount of subsequent processing. A threshold selection method for edge gradient image segmentation based on Otsu method is proposed in order to segment the road edge image effectively under uneven illumination, combining the geometric characteristics of lane distribution on the road surface, Hough voting results. Based on the correlation between road images and the characteristics of lane line width, a candidate lane selection and scorecard algorithm is proposed to identify lane lines in multi-lane scene, and Kalman filter is used to track lane lines. The algorithm software of lane deviation early warning system is used to test and verify. The experimental results show that the total number of frames in road images is 24 661, in which the number of detected frames is 23 483, the number of false frames is 1 178, and the average detection accuracy is 95. 22%. Therefore, the algorithm is correct and effective. It can meet the requirement of real-time and robustness of lane deviation warning system.
【作者单位】：上海大学机电工程与自动化学院;
【基金】：国家自然科学基金项目(61376028) 上海市引进技术的吸收与创新年度计划项目(11XI-15)
【分类号】：U463.6

【相似文献】