基于Beamlet和K-means聚类的车道线识别

Lane Detection Algorithm Based on Beamlet Transformation and K-means Clustering

为了解决视频车道线识别中抗噪性差和鲁棒性低的问题,提出一种基于新的特征提取和分类的快速车道线识别算法。算法首先对预处理后的灰度图像进行改进的Beamlet变换,然后对Beamlet的中点集合运用改进的K-means方法进行聚类分析,最后对每类的中点集合分别进行基于3阶贝塞尔曲线的RANSAC拟合后可以准确地提取出车道线。通过简化Beamlet词典与快速提取Beamlet基,加快了Beamlet变换的计算速度;通过寻找最佳投影线与多次迭代聚类中心来改进K-means聚类,解决了曲线车道线和车道线数目的聚类问题。实验证明,对于结构化或非结构化的道路环境,提出的算法都具有很好的可靠性、实时性和鲁棒性。

In order to solve the problems of the poor noise immunity and low robustness in the video lane detection,a fast lane detection method based on new feature extraction and classification was proposed. The set of midpoints of optimal beamlets was extracted after using the beamlet transformation in the preprocessed grayscale image,which was applied to the analysis of improved K-means clustering. The points in every class was fitted a straight line or curve by using the RANSAC algorithm based on three orders bezier spline. The simplified beamlet dictionary and extracting beamlet base rapidly greatly speed up the beamlet transform. The search for the best projection lines and the multiple iterations of clustering center were used to improve the Kmeans clustering,which solves the clustering problem of curve lane and the number of lane. The result showed that the algorithm has great reliability,real-time and robustness in the structured or unstructured lane environment.