结合Procrustes分析法和ICP算法的PICP配准算法

PICP Registration Method Based on Procrustes Analysis and ICP Algorithm

为了解决传统ICP算法存在查找最近迭代点较复杂、单向查找导致较多的错误点对、收敛函数易陷入局部最优状况的问题,提出一种基于Procrustes分析对ICP算法进行改进的PICP算法.首先通过比较三维空间8个方向上的初始变换参数和迭代点对距离值寻找出点云数据的最优初始变换参数;然后采用双向查找最近迭代点机制优化ICP算法,并将查找到的点对构成新的点云数据;最后通过Procrustes分析法对点云数据求解最小二乘函数,从而获得较高的配准精度,完成ICP算法的最优收敛.通过牙齿点云数据以及兔子标准数据的配准测试表明,文中采用的算法能够解决尺度变换和非均匀点云配准问题,且配准结果收敛较快,配准误差较小.和传统ICP算法相比,文中的PICP配准算法具有全局收敛性高、迭代次数少、抗噪能力强的优点.

In the paper, a PICP method is proposed to improve an ICP algorithm based on Procrustes analysis, in order to deal with the problems of the traditional ICP algorithm, such as the complication of finding the nearest iteration point, wrong point pairs caused by one way search mechanism, and making the convergence function into local optimum value easily. First, finding the optimal initial transform parameters of point cloud data by comparing the initial transformation parameter and the distance between the two points of iteration point pairsamong eight directions of three dimensional space. Second, finding the nearest iteration points with the bidirectionalsearch mechanism and constituting them into a new point cloud data to optimize the ICP algorithm. Finally,solving the least squares function of the cloud data by the Procrustes analysis, then the optimal convergence ofICP registration algorithm can be obtained with higher registration precision. In the experiments, single toothpoint cloud data and standard rabbit point cloud data are adopted to do the registrations. It shows that the PICPalgorithm can solve the scale-transform and non-uniform point cloud registration problems, with faster convergence and more accurate registration than other methods. Compared with the traditional ICP algorithm, the PICP algorithm shows more benefits in high global convergence, fewer iteration times and strong noise resistance.