An improved iterative closest point algorithm based on the particle filter and K-means clustering for fine model matching

The rigid matching of two geometric clouds is vital in the computer vision and its intelligent applications, such as computational geometry, robotics, shape modelling, surface reconstruction and mapping, and many other fields. The variants of the iterative closest point algorithm were employed as the most noticeable matching algorithm. In traditional ICP algorithms applications for symmetrical geometry matching, the initial uncertainty and the multiple local minima of the distance function adversely affect the alignment process, which leads to weak performance, such as incorrect correspondence, narrow convergence region, and instability. In this study, the novel algorithm fused the ICP algorithm, particle filter and K-means clustering to correctly estimate the transformation ICP parameters. Further guide to initial values of parameters and their covariance obtained by k-means clustering. Then, a particle filter was implemented to estimate accurate values and perform global optimization. In the introduced PF-ICP algorithm, the alignment parameters: rotation angles, scale factor, and translation, were defined as particles elements optimized using a sequential importance resampling (SIR) particle filter. The proposed algorithm was implemented on a medical robot FPGA board and applied to “three symmetrical models” and “noisy and poor datasets.” The calculated variances and estimated parameters were compared with four modified ICP methods. The results show a significantly increasing accuracy and convergence region with an acceptable speed for the practical conditions.