Research On Object Artifact Identification And Location Method In Unstructured Environment

In recent years,with the rapid development of the manufacturing industry,the "robots for humans" plan has been gradually implemented,and more and more robots have been applied in industrial sites.Vision-guided robotics has become an important technology in the field of industrial automation.In the unstructured environment,the texture information of objects is small,the mutual occlusion between objects is serious,and the illumination is uneven.The recognition and positioning based on two-dimensional cannot meet the required recognition accuracy and efficiency.Therefore,this paper studies the target workpiece identification and localization method in unstructured environment based on 3D point cloud.The monocular structured light 3D acquisition device is used to obtain the workpiece point cloud data of the stacked scene,and the point cloud preprocessing algorithm and edge extraction algorithm are used to obtain the workpiece without background.The edge features of the workpiece point cloud in the stacked scene of the information,match the target workpiece point cloud with the workpiece point cloud in the stacked scene,and complete the accurate identification and precise positioning of the target workpiece.The specific work of this paper is as follows:(1)Aiming at the problems of disorder,large amount of data,high noise and complex background,which lead to low accuracy and efficiency of recognition and localization of workpiece point cloud data in stacked scenes,an improved preprocessing method for workpiece point clouds in stacked scenes is proposed.This method uses the KD tree space index structure to construct the topological relationship of the workpiece point cloud in the stacked scene,and realizes the rapid search of the nearest neighbor range.Combined with the pass-through filtering algorithm and the voxel grid downsampling filtering algorithm,the simplification of the workpiece point cloud data in the stacked scene is realized.and noise suppression;an accelerated Euclidean clustering algorithm based on KD tree is given to extract the workpiece point cloud of the stack scene without background information;the principal component analysis method is used to calculate the surface normal vector of the workpiece point cloud of the stack scene.(2)Aiming at the problems of low recognition efficiency and inaccurate positioning caused by the repeated features of flat point pairs on the surface of flat workpieces,an improved edge extraction algorithm for workpiece point clouds in stacked scenes based on Gaussian clustering is proposed.The principal component analysis method is used to reduce the data dimension,and the eigenvalue analysis method is used instead of the Gaussian clustering algorithm to identify the edge features of the 3D point cloud,which effectively improves the extraction efficiency of the edge feature of the workpiece point cloud in the stacked scene.(3)Aiming at the difficulty of recognition and localization caused by the stacking and mutual occlusion of target workpieces in an unstructured environment,a global modeling and local matching recognition and localization algorithm is proposed.In the offline phase,the Point Pair Feature is calculated and stored in the hash table as a key value,and a global model description is established.In the online stage,a scheme based on generalized Hough voting is used to obtain a series of candidate poses of the target workpiece,and the k-meansbased clustering algorithm is used to derive the initial pose transformation relationship from the candidate pose clustering.Finally,point to plane is proposed.The ICP algorithm optimizes and iterates the initial pose transformation relationship to obtain the precise 6D pose of the target workpiece in the stacked scene.(4)In order to verify the correctness and effectiveness of the algorithm proposed in this paper,an unstructured environment system experimental platform is built,and experiments on target workpiece recognition and positioning in a variety of complex scenarios are carried out,and the feasibility of the algorithm is verified through robot grasping experiments.The experimental results show that the algorithm proposed in this paper meets the requirements of industrial production in terms of accuracy and efficiency.