基于佳点集人工鱼群的点云配准算法

针对点云配准迭代最近点(iterative closest point, ICP)算法对点云的初始位置姿态有较高的要求且易陷入局部最优的问题, 文章提出一种基于佳点集人工鱼群的点云配准算法。首先采用佳点集方法对人工鱼群初始化, 解决人工鱼群因初始种群分布不均而陷入局部最优的问题, 并通过下采样与三维尺度不变特征变换(3D scale invariant feature transform, 3D SIFT)特征点提取简化点云; 然后采用快速点特征直方图(fast point feature histogram, FPFH)特征描述解求点云间的对应点对并剔除错误对应点对, 通过佳点集人工鱼群算法寻优刚性变换的6个参数完成粗配准; 最后使用ICP算法完成精配准。实验选取斯坦福大学提供的Bunny、Dragon和Happy Buddha 3组测试数据集进行配准; 结果表明, 该文算法收敛速度快, 能为ICP算法提供良好的初始位姿避免其陷入局部最优。

Aiming at the problem that the iterative closest point (ICP) algorithm has higher requirements for the initial position and posture of the point cloud and is easy to fall into the local optimum, a point cloud registration method based on artificial fish swarm algorithm (AFSA) using good point set (GPS) is proposed. In order to prevent the artificial fish swarm from falling into a local optimum due to the uneven distribution of the initial population, the algorithm first uses the GPS method to initialize the artificial fish swarm. The point cloud is simplified by down-sampling and 3D scale invariant feature transform (3D SIFT) feature point extraction algorithm. Then, the corresponding point pairs between the point clouds are obtained by the fast point feature histogram (FPFH) feature description and the wrong corresponding point pairs are eliminated, and the six parameters of the rigid transformation are optimized by the GPS based AFSA to complete the rough registration. Finally, the fine registration is completed by the ICP algorithm. Experiments on Bunny, Dragon and Happy Buddha models of Stanford University show that the algorithm converges fast and can provide a good initial pose for the ICP to prevent it from falling into a local optimum.