3D workpiece pose estimation based on adaptive subpattern manifold learning
1. School of Agricultural Equipment Engineering, Jiangsu University, Zhenjiang, Jiangsu 212013, China; 2. School of Aviation Engineering, Jiangsu Aviation Technical College, Zhenjiang, Jiangsu 212134, China
Abstract：To locate threedimensional workpiece of monocular vision timely and accurately in complex industrial environment, a pose estimation method of threedimensional workpiece was proposed based on adaptive subpattern manifold learning (SPIVP). The constructing manifold method was given by nonlinear reduction of dimension framework and reconstruction of high dimension space, and the low dimensional feature subspace was obtained to maintain the optimal continuity of nature variable. The pose estimation of workpiece was realized based on the manifold construction method. The pose estimation method of workpiece with occlusion was proposed based on SPIVP after the segmentation rules of adaptive subpattern was given. Three kinds of common workpieces were tested, and the horizontal rotation and the vertical rotation were chosen as natural variables to conduct pose estimation of workpieces with or without occlusion. The results show that the average pose estimation time of the proposed method is 73.6 ms, which can meet the requirement of realtime processing. The positioning accuracy rates of screwdriver, crankshaft and cylinder are 95.4％, 96.1％ and 98.4％, respectively. The recognition accuracy of the proposed method is higher than those of other methods in different occlusion cases. The subpattern segmentation method is performed with higher recognition rate than the method without subpattern segmentation.
2018, Vol. 39 
