Point Cloud Registration Algorithm Based on Fusion of FPFH and SHOT Features

Point cloud registration is a crucial step in three-dimensional data processing, and the Iterative Closest Point (ICP) algorithm has been widely applied to this task. To address the sensitivity to initial pose and the tendency to become trapped in local optima inherent in traditional ICP, and to improve the speed and accuracy of point cloud registration, a feature-integrated point cloud registration method is designed. First, the coarse registration method is optimized. For feature point extraction, point cloud downsampling is performed using the voxel grid method, which maintains the features of the original point cloud well while reducing the amount of data to be computed. Secondly, feature values obtained from Fast Point Feature Histograms (FPFH) and Signature of Histograms of Orientations (SHOT) feature descriptors are fused proportionally. Dynamic weights are calculated to automatically adjust based on the feature response values, ensuring optimized fusion performance in different scenarios. Subsequently, the Sample Consensus Initial Alignment (SAC-IA) algorithm is combined to perform coarse registration on the point cloud. The fused features can provide more accurate correspondences to reduce the number of iterations, thereby obtaining a relatively accurate initial transformation matrix. In the precise registration stage, the availability of a good initial pose further improves the registration accuracy and speed of the algorithm. Finally, comparative experiments are conducted to verify the effectiveness of the proposed registration algorithm. The experimental results demonstrate that the proposed registration algorithm exhibits faster convergence and excellent registration accuracy, indicating that the proposed registration algorithm is effective and meets the requirements of three-dimensional point cloud pose measurement.