Research On Visual Recognition Technology And Its Application For Intelligent Manufacturing

At present,a new round of technological innovation with intelligent manufacturing as the core is affecting the new pattern of the world’s manufacturing industry in the future.In order to seize the opportunity of this round of emerging technology development and stabilize the leading position of domestic manufacturing industry,western developed countries have put forward development strategies for domestic manufacturing industry one after another.Only by accelerating the breakthrough and wide application of key technologies of intelligent manufacturing,can we promote the transformation and upgrading of China’s traditional manufacturing industry and narrow the gap between China and developed countries.In order to effectively promote the application of intelligent manufacturing,the current implementation of the "machine replacement" work through the "equipment + robot" way to replace the traditional way of artificial manufacturing.Machine vision is the "smart eye" that can improve the intelligent level of robot.This paper will carry out the research on Key Technologies of robot vision grasping,including camera calibration and hand eye calibration technology,workpiece recognition method,workpiece positioning and pose detection method,and the experimental verification of workpiece recognition and positioning effect applied to the production line.Firstly,this paper constructs a monocular vision positioning system,and carries on the theoretical research and process derivation of monocular vision camera calibration.Secondly,the convolution neural network theory is studied,and the recognition of single target workpiece is completed based on vgg16 model,the test accuracy is 100%.Then,the basic principle of Yolo model is studied,and the recognition and location of multi-target workpiece is completed based on yolov4 model.The accuracy of verification set is 0.99,which can meet the real-time recognition and location requirements of multi-target workpiece.Then,the image preprocessing methods are studied,and the advanta ges and disadvantages of several filtering methods are compared.Finally,the Gaussian filter is selected to denoise the image,and the open operation and closed operation in morphological transformation are used to further preprocess the image.Based on t he characteristic moment calculation in opencv,the workpiece centroid and minimum circumscribed rectangle are captured,The coordinate position and pose angle of the workpiece are obtained.Finally,it is applied to the actual production line to test the recognition and positioning effect of the algorithm model.The results show that the model can recognize and locate multiple objects.When there are interference parts,the interference parts can be eliminated,which can meet the demand of real-time detection of production line workpiece.