基于PCA和Delaunay剖分点云配准算法研究

Study on Coarse Registration of Point Cloud Based on PCA and Delaunay Triangulations

针对三维点云精配准算法无法处理初始位置相差较大的点云并易陷入局部最优的问题,提出一种基于特征值计算和匹配的粗配准算法。由于点云表征对象表面形状的一致性,通过主成分分析(PCA)法可以得到每个点的特征值,通过构建基于特征值的Delaunay三角网可以加速寻找特征值最接近的点,利用随机抽样一致性(RANSAC)算法可以获取优化的变换参数。实验表明,该粗配准算法可以有效处理点云初始位置较差的情况,将两点云调整到较好的位置,保证了大部分区域的重叠。

The refined registration algorithm can get into local optimum easily when two 3D point cloud has large position difference.Aiming at this problem,a novel coarse registration algorithm based on computing and matching eigenvalue is proposed.The eigenvalue in the same position of two arbitrary point clouds is represented consistently.The eigenvalue in every point can be achieved by principal component analysis algorithm.Then,Delaunay triangulations of model point cloud are constructed in the process of searching closest point in order to accelerate the iteration.In the last step,RANSAC algorithm is utilized to realize the optimal transformation.The experiment shows the improved registration algorithm can adj ust one of the two original point cloud that is located in its own coordinate system and far away from the other effectively so that they can be aligned.The following registration algorithm can be j ustly applied to refining the previous result.