Abstract:
To address the susceptibility of clustered-hole inspection on revolved surfaces to projection distortion and the associated computational cost in 3D point-cloud measurement, a multi-type clustered-hole detection method based on dimensionality-reduced unwrapping of 3D point clouds is proposed. First, a cylinder is fitted by the least-squares algorithm, and the 3D point cloud is unwrapped onto a 2D plane, whereby the inspection task is converted into an efficient image-processing problem. Subsequently, the 2D point cloud is rasterized, and gaps are repaired by morphological closing, so that hole-boundary contours can be extracted accurately. Finally, a multi-feature fusion classification algorithm based on circularity, angle, and area ratio is presented, by which circular, rectangular, and other irregular holes are identified automatically and their dimensions are extracted. Experimental results demonstrate that recognition accuracies of 99.44% and 100% are obtained for circular and rectangular holes, respectively, corresponding to an overall accuracy of 99.46%. The overall root-mean-square errors are 0.056 0 mm and 0.081 3 mm, respectively. Both the efficiency and the accuracy of clustered-hole detection are thereby significantly improved.