| Machine vision is a hot topic in the field of mechanical manufacturing and automation.It is the development trend of intelligent manufacturing to apply machine vision to i ndustrial robots,to realize the detection,segmentation and tracking of moving targets.Object segmentation is the separation of objects from image or video frames from background.It is the premise for subsequent target recognition and tracking.The exis ting industrial visual detection platform is usually used as a global or local threshold segmentation algorithm.The segmentation precision is not ideal in the dynamic scenes such as complex background or light shadow change.In this paper,an improved target segmentation method is proposed on the basis of fast target segmentation algorithm,which combines the idea of background learning and improves the segmentation effect by training the background model.Further combining the full convolution neural netw ork deep learning model,we achieve the high-precision segmentation of workpiece targets in dynamic scenes by learning the workpiece images.The main contents of this paper are as follows:First of all,this experiment platform is introduced,which is the visual inspection platform for industrial robots,in which the selection rules of industrial cameras and image acquisition card are explained in detail.Secondly,the classical algorithm of target segmentation,the existing target segmentation algorithm of the industrial robot platform and the basic principle and the implementation steps of the fast moving target segmentation method under the unrestricted scene are introduced.The above algorithms are used respectively in the image data obtained on the industrial robot visual detection platform,and the experimental results are added to the experimental data.The comparison and analysis were performed.Thirdly,the basic concepts and modeling methods of the mixed Gauss model are introduced.On the basis of t he fast segmentation method of moving target in the non restricted scene,a mixed Gauss model is proposed to learn the background of the conveyor belt,and the rod segmentation of the moving object on the industrial robot platform is realized.The experimental results are compared with those of fast moving object segmentation method under unrestricted scenes.Finally,the total convolution neural network training method is applied to the dynamic scene workpiece target segmentation based on the industrial robot vision inspection platform.First,the structure and training methods of the full convolution neural network(FCN)model are introduced.Then,the FCN model is trained by 1000 moving workpiece images and the test results are obtained.The high precisio n segmentation of the workpiece target in high dynamic scene is realized.Finally,the experimental results of the fast segmentation method and the fast segmentation method based on the background learning are compared and analyzed with the threshold method and the unrestricted scene.This paper applies the fast target segmentation method under the unrestricted scene to the dynamic workpiece segmentation on the industrial robot platform.On the basis of this method,the method of background learning is proposed to improve the segmentation precision,and the high precision segmentation of the workpiece target in dynamic scene is realized in combination with the model training method of the full convolution neural network.It provides a method and basis for so lving the accurate location of industrial robot vision system in complex scenes. |
