To overcome the limitations of the traditional random sample consensus (RANSAC) algorithm in cylindrical segmentation, a novel method is developed for segmenting point clouds of ring-ribbed shells by integrating structural features and statistical methods.

Initially, the model surface area feature is utilized to estimate the proportion of inliers, thereby enhancing the accuracy of initial parameters. Subsequently, principal component and radius constraints are introduced to enhance the accuracy of cylinder identification and reduce the number of iterations. Then, a weight function-based correction method is applied to mitigate outlier interference, thereby improving the accuracy of cylinder fitting. Finally, the DBSCAN algorithm clustered the point clouds of ring-ribs, and an improved RANSAC algorithm identified localized features, thus achieving precise measurement of component dimensions.

Experimental results show that the proposed method effectively addresses the intelligent recognition and dimensions measurement of components in various parts of the ring-ribs, significantly improving the recognition speed and accuracy of cylindrical shell and ring-ribs. The precision, recall, and overall accuracy of cylindrical shell reach 96.9%, 99.5% and 96.4% respectively, with a computational speed increase of approximately 4.6 times. The measurement error for ring-rib component dimensions is within 0.2%.

Compared with traditional methods, the proposed method offers significant advantages in the accuracy and computational efficiency of point cloud segmentation.