The existing filtering methods for airborne LiDAR point cloud have low accuracy. An adaptive filtering algorithm is proposed which is improved based on multilevel resolution algorithm. First double index structure of Octree and KDtree is established. Then the initial reference surface is constructed by ground seed points. According to the slope fluctuation situation, the grid resolution of the ground referential surface is adjusted in an adaptive way. Finally, the refined surface is formed gradually by multilevel renewing resolution to provide filtered point cloud with high accuracy. Experimental results show that the error of Type II can be effectively reduced, the average Kappa coefficient increases by 0.53% and the average total error decreases by 0.44% compared with multiresolution hierarchical classification algorithm. The result tested by practically measured data shows that Kappa coefficient can reach 90%. Especially, it maintains advantages of high accuracy under complex topographic environment.

现有的机载LiDAR点云滤波方法精度不高。提出了一种基于多级分辨率算法改进的自适应滤波算法。第一个Octree和KDtree的双索引结构建立。然后由地面种子点构建初始参考面。根据坡度波动情况，自适应调整地面参考面的网格分辨率。最后，通过多级更新分辨率逐渐形成细化表面，以提供高精度的过滤点云。实验结果表明，与多分辨率分层分类算法相比，可以有效降低Type II的误差，平均Kappa系数提高0.53%，平均总误差降低0.44%。经实测数据检验，Kappa系数可达90%。尤其是在复杂地形环境下保持高精度的优势。