摘要
针对弯道和坡道下三维激光雷达地面高准确性提取问题,提出一种基于马尔科夫随机场的道路场景分割算法。利用激光雷达原始数据的数据结构信息,先建立扇形栅格图,利用最小栅格的四邻域栅格的高度和到四个栅格的距离计算最小栅格的梯度值;建立以最小栅格为节点的马尔科夫随机场。通过最大流/最小割算法求解马尔科夫无向图模型,对所有栅格进行分类,标记为路面点和非路面点。算法分别在校园路况和城市公路路况下进行了试验,并且与高度阈值算法进行对比,结果表明:在弯道和坡道相较于高度阈值算法有较好的鲁棒性。
An algorithm with Markov Random Field-based road scene segmentation is presented to play a road scene with curves and slopes from 3 D Lidar data with high quality.Based on the structure data information of the original Lidar data,a fan-shaped grid map is established.The gradient of the smallest grid is calculated by utilizing the corresponding height and distance with their four neighborhoods;Markov Random Field in node with the minimum grid has been built.Solving undirected graphical model in maximal flow/minimal cut,all grids has been categorized as marked road and non-road points.Experiments are taken under both campus conditions and city conditions.The results demonstrate that the proposed algorithm has better robustness than the height threshold algorithm in the road scene with curves and slopes.
作者
邹斌
冯昊文
刘康
ZOU Bin;FENG Haowen;LIU Kang(Hubei Province Key Laboratory of Modern Automotive Technology,Wuhan University of Technology,Wuhan 430070,China;Hubei Collaborative Innovation Center for Automotive Components Technology,Wuhan University of Technology,Wuhan 430070,China)
基金
湖北省科技厅资助项目(2016BEC116,2017BEC196)
