基于分层粒子群优化的三维点云配准

Three-dimensional Point Cloud Registration Based on Hierarchical Particle Swarm Optimization

为了提高三维点云配准的性能,采用基于分层粒子群优化的迭代最近点算法来完成点云配准;首先将源点云作为粒子群粒子,将粒子分成多个子群,然后以点云的曲率为适应度值,分别求解子群适应度值和全局粒子适应度值,并将子群适应度值、全局粒子适应度值和粒子当前速度三者结合,共同搜寻最优粒子,以得到能够精确表达点云结构的特征点,最后采用迭代最近点算法对特征点进行配准。仿真结果表明,通过合理设置粒子速度权重和子群规模,相对于标准迭代最近点算法,分层粒子群优化算法的三维点云配准效率提升显著,配准均方误差略有减小。

To improve performance of three-dimensional point cloud registration,iterative closest point algorithm based on hierarchical particle swarm optimization was used to complete the point cloud registration.Firstly,source point clouds were regarded as particles in the particle swarm,and the particles were divided into several subgroups.Curvature values of point clouds were then taken as fitness values to solve the subgroup fitness values and global particle fitness values respectively.The subgroup fitness values,global particle fitness values,and current speed of particles were combined to search for the optimal particle,so as to obtain feature points which could accurately express point cloud structure.Finally,iterative closest point algorithm was used for registration of the feature points.The simulation results show that by reasonably setting weights of particle velocity and subgroup size,the efficiency of three-dimensional point cloud registration of hierarchical particle swarm optimization algorithm is significantly improved and the registration mean square error is slightly reduced compared with standard iterative closest point algorithm.