基于车载连续序列图像的道路曲率计算方法

提出一种基于车载连续序列图像的道路曲率计算方法.算法首先利用车载视 觉系统进行内外参数标定实现对路面车道线的三维重建，利用三维重建误差模型对重建 的数据点进行误差估计，并对数据点进行选弃.对序列图像中路面车道线重建出来的曲线 采用ICP 算法进行曲线匹配，并将匹配后的曲线映射到统一的参考坐标系中.最后利用匹 配曲线上的数据点进行圆周拟合，计算道路曲率.实验中利用真实的车载连续序列图像对 算法进行验证.针对曲率半径分别为96 m和430 m两段不同程度的弯道，利用本文算法计 算对应道路的曲率半径.结果表明，本文算法能够精确地计算出不同弯道的道路曲率,特 别是针对平缓弯道的曲率计算，比基于单幅图像的曲率计算方法更加准确与稳定.

Computation of Road Curvature from a Sequence of Consecutive In-vehicle Images

A novel method to compute road curvature is proposed by using a sequence of consecutive log images. We first calibrate the in- vehicle camera system and reconstruct road lane lines. The error analysis model is applied to evaluate the reconstructed 3D data points such that only those data points with low reconstruction errors are utilized for curvature computation. The reconstructed curves from consecutive log images are registered with Iterative Closest Point (ICP) algorithm and mapped into a reference coordinate system. The 3D data on the mapped curves are finally used for circle fitting to compute the radius and the curvature. The proposed curvature computation method is tested with real video log images on two different road segments with the corresponding radii of 96 m and 430 m. The computation results demonstrate that the proposed method is both accurate and practical to compute the curvature. Furthermore, it outperforms the single-image based method for slow curvature computation.