一种实时车道线偏离预警系统算法设计和实现

Algorithm design and implementation for a real-time lane departure pre-warning system

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。

Aiming at the real-time lane departure pre-warning question, a lateral erosion operator was used to corrode the image after edge detection, and the irrelevant edge information in image was reduced and eliminated to decrease the follow-up processing data quantity significantly. A threshold selection method of edge gradient image block based on Otsu algorithm was proposed to effectively partition road edge image under asymmetrical illumination. A lane line voting selection and scoring algorithm combining the geometrical characteristics of lane line distribution on road, Hough voting results, the correlation of road images and the width characteristics of lane line was proposed to recognize the lane line in multilane scenes. Kalman filter method was applied to track the lane line. The algorithm software of lane departure pre-warning system was used to carry out the test verification. Test result shows that total frame number of road images is 24 661, the frame number of right detection is 23 483, and the frame number of wrong detection is 1178, and the average detection accuracy is 95.22%. The test verified the correctness and effectiveness of the algorithm, which can satisfy the real-time feature and robustness of lane departure pre- warning system. 1 tab, 9 figs, 27 refs.