Abstract
This study introduces an innovative feature point extraction method combined with an improved Grey Wolf Optimizer (GWO)-based coarse registration approach to address common challenges of low registration accuracy and slow processing speed in point cloud registration. The feature extraction design method begins by projecting the point cloud onto a uniformly segmented sphere. Principal component analysis (PCA) is then employed to compute the curvature change rate of the point set within each patch area. Subsequently, sampling weights are assigned nonlinearly based on the calculated change rates, facilitating effective feature point extraction. The extracted feature points serve as the initial values for the improved gray wolf optimization algorithm, which is employed to refine the registration results. Experimental comparisons conducted on three public datasets demonstrate that the feature extraction method proposed in this study achieves improved accuracy and efficiency. Furthermore, the registration results substantiate that our method outperforms other algorithms with respect to both accuracy and computational efficiency.