Enhanced iteration closest point is based on stochastic differential evolution algorithm and random sampling

To address the issue of low accuracy in 3D point cloud registration, we present a novel iterative closest point (ICP) algorithm for point registration using the stochastic differential evolution algorithm and random sampling (SDE_ICP). Given ICP's sensitivity to the initial transformation matrix, our approach divides the algorithm into two stages: coarse registration and fine registration. In the coarse registration phase, we introduce a random sampling algorithm to the original point cloud, expediting the coarse registration process. Subsequently, we propose a stochastic differential evolution (SDE) algorithm with random opposition-based learning population initialization, random centroid mutation, and a boundary control strategy. The search individual comprises translation and rotation transformation parameters, offering a robust relative initial position for subsequent ICP fine registration. Finally, we conduct simulations and analyses using Bimba, Angle, and Bunny datasets from FGR to evaluate SDE_ICP. We compare its performance with ICP, TrICP, DE_ICP, and SaDE_ICP. The results indicate that the proposed method exhibits superior accuracy.