Research On Workpiece Recognition And Location Technology Based On Machine Vision

Product assembly is an indispensable link in the production process,which needs to occupy most of the time of the whole product processing.At the same time,the performance of the product has a direct relationship with the assembly accuracy.For the small batch customized product line,higher requirements are put forward for the rapidity and accuracy of assembly.Due to the continuous development and progress of numerical control technology and sensors,automatic assembly based on robot can greatly improve the production efficiency and shorten the production cycle.At present,the industrial robot can not achieve complete automatic assembly,for some specific occasions still need auxiliary facilities to cooperate.The recognition and location of workpiece is the main technical difficulty in automatic grasping.Aiming at this problem,this paper proposes a method of workpiece recognition and location based on machine vision.The contents are as follows:Firstly,the overall scheme design of target workpiece from two-dimensional image recognition to three-dimensional positioning is proposed.The technical background and significance of workpiece recognition and positioning based on machine vision are introduced.According to the actual engineering requirements of the workpiece,the recognition and positioning method of the workpiece based on binocular vision is designed.Secondly,an improved object detection algorithm based on SSD is designed.Aiming at the problem of low detection accuracy of small target workpieces in SSD,based on SSD algorithm,the backbone network vgg16 of SSD is replaced by resnet50 to expand the number of network layers with different sizes.At the same time,the algorithm idea of feature pyramid(FPN)is added to fuse the low-level features and high-level features in the network to solve the problem of insufficient ability of extracting low-level features in SSD algorithm,So the defect of poor detection effect for small target workpiece is improved.The experimental results show that the improved SSD workpiece target detection method reduces the miss detection rate of small targets,and improves the overall detection accuracy of the workpiece.Then,the process from 2D image recognition to local 3D point cloud reconstruction is realized.The two-dimensional image detected by SSD is segmented,and the two-dimensional image of a single target workpiece is segmented to establish a binocular camera model.The internal parameters of the camera are obtained by camera calibration method.Through the pixel matching of the same name points,the 3D reconstruction process of the target workpiece is realized,and the 3D point cloud of the workpiece is obtained.Finally,a point cloud registration algorithm based on PCA is proposed.In order to get the position and position of the workpiece,the principal component analysis method is used to get the main axis direction of two groups of workpiece point clouds,and the initial transformation matrix of the point cloud is obtained.The error analysis method is used to correct the initial registration error and solve the problem of the spindle reverse.Finally,the precise registration process of the workpiece point cloud is realized by ICP algorithm,and the final position and attitude of the target workpiece are obtained.