Identification Of Component And Its Type Under Different Attitude Based On SVM And Parzen Window

With the development of industrial robots technology,relying on visual guidance,positioning and identification technology is increasingly important.The visual recognition of the workpiece and the identification of the workpiece type are the important steps to realize the robot grasp the workpiece.How to recognize and classify the workpiece after attitude transformation has important theoretical and engineering research value.According to the proposed workpiece invariant moment feature based on Support Vector Machine(SVM)workpiece visual identification research scheme,an experiment platform of workpiece attitude transformation is built in the laboratory environment,and the visual recognition of workpiece attitude transformation is experimentally studied.The main research contents are listed as follows:Firstly,a 6-DOF attitude-change experiment platform is designed and analyzed,and the functional parts of each part of the platform are selected and analyzed.The 6-DOF platform structure is combined with the vision system to collect the workpiece images under different attitude transformations.The experimental platform is assembled to meet the design performance requirements.Secondly,image filtering,image enhancement and watershed image segmentation are analyzed,which can lay the foundation for the extraction of invariant moments of the workpiece image.Thirdly,a single workpiece,partially occluded workpiece and partially overlapped workpiece are selected for image acquisition and workpiece samples are constructed,and seven feature moment invariants are extracted for the workpiece rotation in the horizontal plane and in the three-dimensional attitude transformation.Finally,a regression model of workpiece identification based on the SVM is established.The visual recognition of workpiece is realized by combining the moment invariant feature data with SVM.In order to further realize the recognition of workpiece types,the combination of Parzen window and particle swarm is used to realize the average of the fifth and seventh order moments of the workpiece training samples taken as the object classification target.And the Parzen window estimation result is optimized by the particle swarm optimization with inertia.At last,the discrimination of the workpiece type is realized.