对点云预处理,并采用点云映射的方式快速分割出地面,同时消除路内障碍物以降低数据量;将分割出的地面数据组织成无向图,结合道路边界的多种局部特征和全局连续性特征提取边界点;根据道路边界点的测量模型修正提取的边界点,并采用二次多项式拟合修正后的边界点;采用多种策略对道路边界进行更新以使相邻两帧检测的道路边界保持平滑.实验证明,在道路边界不规则、存在路内障碍物遮挡边界的情况下,采用该方法得到的道路边界检测结果依然具有较高的鲁棒性和准确性.%The point cloud was preprocessed and used to segment the ground part from the background rapidly based on point mapping.Simultaneously,with the preprocessed points cloud,the obstacles in the road can be removed to reduce the number of points and refine the input data for the coming process.The segmented ground data was organized into an undirected graph.With fusing several local features and global continuity features of the road curb,most candidate curb points could be extracted and the fake curb points could be eliminated.Furthermore,a measurement model was utilized to correct the extracted curb points,and then a quadratic polynomial fitting algorithm was applied to get the curve of the two curbs.Finally,several strategies were used to update the detected curbs and smooth the curb curve in the consecutive point cloud frames.The experimental results show that the proposed curb detection algorithm has high robustness and accuracy under the condition of both irregular and occluded road curb even there are obstacles in the road.
展开▼
