Steel Plate Surface Defect Recognition Method Based on Depth Information

Abstract

Although steel surface defect recognition based on 2D image data has been extensively researched over the last ten years, it is very difficult for the identification of the defects with depth information in these methods. This paper presents a recognition method of steel plate surface defect through the estimated 3D depth information. In this method, the 3D data of the steel plate surface are first reconstructed using structure from motion (SFM). Then 3D points of the defect are segmented from the 3D reconstructed result of the steel plate surface using a region-growing based 3D information segmentation method. Finally, normal map is estimated from the segmented 3D point cloud, and the smoothness threshold in the normal map is optimized to classify the defect region and other regions. In experiment, the steel plate specimens with different hole sizes and the non-injured region are prepared, and the defect region based 3D information is classified. Experimental results show that the proposed method is efficient and feasible.