Improved ICP point cloud registration method based on feature point extraction and PCA

Abstract

The traditional Iterative Closest Point (ICP) method for point cloud registration has problems such as poor real-time performance, susceptibility to local extremum, and low registration accuracy. This paper proposes a three-step point cloud registration method based on feature point extraction, Principal Component Analysis (PCA) coarse registration, and ICP fine registration. Firstly, it defines the concept of local density in point cloud data and automatically selects points with higher local density as feature points. Then, it uses PCA to analyze the extracted feature points and calculates the required translation and rotation parameters for registration based on the principal component direction of PCA. Finally, it uses ICP to perform precise data registration. The experimental results show that the proposed method improves registration accuracy by more than 13.4% compared to the comparison methods, improves real-time performance by more than 38.2%, and exhibits higher adaptability under low signal-to-noise ratio conditions, with high application prospects.