基于改进的Hough变换和K-mean聚类的RANSAC车道线检测方法

RANSAC Lane Detection Based on Improved Hough Transform and K-mean Clustering

摘要:

针对车道线检测中特征点匹配方法存在实时性不高和精度低的问题，本文首先提出了基于消失点改进的Hough变换提取特征线，剔除了干扰线，提高的计算量；然后对特征数据集采用 K-means 聚类和RANSAC拟合算法，首先利用 K-means 聚类对改进的Hough变换提取的特征点进行预处理，剔除了孤立的特征点，接着匹配Catmull-Rom 样条曲线进行RANSAC拟合，相当于二次优化，实现了车道线的快速和精确配准。通过实验表明，该算法不仅提高了车道线识别的精确度，而且具有很好的鲁棒性。

Abstract:

In order to solve the problem of feature point matching in lane detection with low real-time accuracy and low accuracy, this paper first proposes an improved Hough transform based on the vanishing point to extract the characteristic line, eliminates the interference line and improves the computational complexity. K-means clustering and RANSAC fitting algorithm were used in the dataset. Firstly, K-means clustering was used to preprocess the feature points extracted from the improved Hough transform, and the isolated feature points were removed, and then matched with Catmull-Rom spline curve RANSAC fitting, which is equivalent to quadratic optimization, enables fast and accurate registration of lane lines. Experiments show that this algorithm not only improves the accuracy of lane line recognition, but also has good robustness.