基于目标结构和运动的船舶激光点云配准算法

Ship laser point could registration algorithm based on structural features and motion

由于在野外环境下船舶结构特点及激光雷达测量视野限制,所获取的点云图像存在特征匹配错位和缺失现象,导致经典的配准算法精度较低。在分析点云结构特点和目标运动轨迹的基础上,提出一种基于目标结构和运动轨迹的配准算法。在对待配准点云进行预处理后,通过结构特征的引入解决了迭代最近点法(Iterative Closest Point,ICP)算法特征匹配错位陷入局部最优解的问题;在此基础上利用得出的主体变换矩阵求解出目标的运动轨迹,并根据运动轨迹对缺乏有效特征的点云进行配准同时校准主体点云。实验结果表明,以米为点云坐标单位,相比于ICP和4PCS+ICP方法,所提出方法在船舶点云配准上的均方误差(Mean Square Error,MSE)降低了0.2左右。

Due to the structure of the ship in the wildand the limitations of the LiDAR measurement field of view,the acquired point clouds suffer from misalignment and lack of features,which leads to the low accuracy of the classical registration algorithms.Based on the analysis ofthe structural features of the point clouds and the motion trails of the ships,an alignment algorithm based on structural features and motion trails is proposed.After pre-processing the point clouds to be aligned,the structural features are introduced to solve the local optimal solutions of iterative closest point(ICP)algorithm caused by the misalignment of features.On this basis,the motion trails of the ships are solved by the transformation matrix of the main part of point clouds.Based on the motion trails,the point clouds lacking of effective features are registered and the registrations of the main part of ship point clouds are calibrated.The experimental resultsshow that coordinating in meters,the mean square error(MSE)of the proposed registration algorithm is 0.2 lower than ICP and 4PCS+ICP algorithms.