一种基于随机抽样一致性的车道线快速识别算法

针对现有车道线识别算法的有效性、实时性和鲁棒性不高的问题,提出了一种改进的快速随机抽样一致性(RANSAC)的曲线拟合验证的视觉车道线识别算法.该算法首先在进行逆透视变换后选用各向异性的高斯核滤波;然后对不同光照亮度图像采用适应性强的分位数方法进行二值化,并针对车道线在变换图中几乎垂直的特性,再利用直方图统计法检测出初始车道线;最后用改进的快速RANSAC的曲线拟合算法进行曲线修正,找出车道线可能存在的弧度,使检测的曲线更加精确.为提高检测的精度,最后对识别结果进行后处理.实验结果证明,对各种复杂的城市道路,所提出的算法均具有很高的鲁棒性和有效性,且算法处理效率很高,能很好地满足智能车实时检测车道线的要求.

A Fast Algorithm Based on RANSAC for Vision Lane Detection

Abstract： In view of the problems that the real-time, robustness and efficient of the existing lane detection algorithm are low, an improved and fast vision lane detection algorithm based on RANSAC (random sample consensus) was proposed. First, the inverse perspective mapping was conducted. Then, the image was filtered using anisotropic Gasssian filters. The quantile threshold method which has a strong adaptability to different illumination brightness image was used to the filtered image. The initial lines were detected using the histogram statistics method because almost all the lanes in the transform image were vertical. After that, an improved and fast RANSAC curve fitting step was performed to refine the detected initial lines and correctly detect curved lanes. Finally, a post processing was conducted to further improve the accuracy of algorithm. The results show that the improved algorithm has a great robustness, strong stability and high efficiency, which can meet the requirements of intelligent vehicle real time detection.
Key words：