Novel Range Image Segmentation Using Region-Growing and Surface Classification

In the article, a novel method for object segmentation using surface characteristics of range images is presented to solve the problem of digital background identification (DBI) in 3-D measurement. The proposed method uses a novel criterion based on the distribution of normal surface vectors for 3-D surface (or object) segmentation. According to this criterion, scanned range data from measured objects is first classified into several types of surface as an initial stage of evaluation for addressing all the measured points belonging to the imaging background. By incorporating this criterion into the region-growing process, a robust range-data segmentation algorithm capable of segmenting complex objects, which may be suffered with huge amount of noises from a real field condition, is established. Furthermore, to detect the object boundary accurately, a recursive search process involving the region-growing algorithm for registering homogeneous surface regions is developed. Experiments from several scenarios using a laser 3-D scanner demonstrate the feasibility and effectiveness of the proposed method for 3-D object segmentation and localization in automated optical inspection (A0I).