基于消失点迭代重估的道路边缘检测

Road Edge Detection Based on Vanishing Point Iteration Revaluation

传统道路边缘检测算法很少能用于乡村道路。该文提出基于消失点迭代重估的道路边缘检测方法:Gabor滤波器估计像素纹理方向,同时计算对应的置信度;根据纹理方向,由置信度大于阈值的像素投票确定初始消失点;以初始消失点为起点,向下建立一组射线,并计算由颜色差和二倍角正弦函数加权的方向一致率,将最大值对应的射线作为初始道路边缘;交替采样两侧边缘,选取最优消失点,由此得到两条道路边缘。实验测试了不同类型的乡村道路图片,并与经典算法对比。结果表明该方法在乡村道路中具有较高精准度与鲁棒性。

Few traditional road edge detection algorithms can be employed in rural road. An edge detection method based on vanishing point iteration revaluation is presented in this paper. Firstly, the texture orientation of pixel is estimated by Gabor filter, and the corresponding confidence is computed. Secondly, according to the texture orientation, the initial vanishing point is voted by pixels which their confidence are greater than threshold. Thirdly, with the initial vanishing point as starting point, a group of imaginary downward rays are founded, then the orientation consistency ratio weighted by both color difference and double-angle sine function is calculated, and the ray with maximum value is selected as the first road edge. Finally, the optimum vanishing point is given by sampling bilateral edges alternately, and thus two road edges are obtained. A variety of rural road pictures are tested by experiment, and some comparisons are made with the traditional algorithm. The results indicate the method is accurate and robust in rural road.