Enhancing Point Cloud Registration Precision of Conical Shells Through Edge Detection Using PCA and Wavelet Transform

Reliability assessment of conical shells in the chemical industry commonly relies on point cloud registration. Thus, accurate edge detection from 3D laser scan data is crucial for high-precision registration. However, existing edge detection methods often misclassify or omit gradual edge points on conical shell structures, significantly compromising registration accuracy and subsequent integrity assessment. This paper proposes an edge point detection method integrating Principal Component Analysis (PCA) and wavelet transform. First, characteristic curves are constructed by computing the ratio of PCA eigenvalues at all points to generate preliminary candidates for gradual edge points. Subsequently, distance vectors are calculated between the centroid of each characteristic curve and its sampled points. These vectors are then encoded via multi-level wavelet transform to produce mapping vectors that capture curvature variations. Finally, gradual edge points are discriminated effectively using these mapping vectors. Experimental results demonstrate that the proposed method achieves superior edge detection performance on complex conical shell surfaces and significantly enhances the accuracy of point cloud registration.