Semantic feature-based point cloud segmentation method for tire tread

This study addresses the challenge of accurately segmenting complex tire tread patterns in point clouds using a novel method based on semantic features. This approach determines the segmentation strategy by considering the tread semantic attributes and the growth of the pattern boundary point cloud values within the K-neighborhood range. Initially, the 3D tire tread point clouds are transformed into 2D point cloud mapping matrices through uniform discrete sampling and mapping techniques. Then, principal component analysis and cubic spline interpolation are employed to segment pattern groove features by following the pattern growth law and adopting the region-growing method. Intricate features, such as sipes, are segmented by analyzing morphological features and concavity curvature within the K-neighborhood. Finally, agglomerative hierarchical clustering algorithms are utilized for denoising and boundary refinement. Experimental results demonstrate the method's precision, achieving a mean intersection over union ratio of 92.61 % and a maximum average centroid deviation of 0.112 mm.