Research On Cuboid Like Objects Detecting System Based On Machine Vision Recognition

With the development of intelligent manufacturing technology,machine vision is more and more widely used in life and industry.In order to solve the classification problem of cuboid like objects in life and industry,this paper takes the medicine bottle sorting as an example,and combines robot technology and convolutional neural network to sort the medicine bottles.The medicine bottle image is collected by CCD camera,and the image is trained by the improved VGG network,the recognition result is sent to the robot,Then robot adopts the grabbing strategy according to the position and category information of the medicine bottle.this paper,aiming at the problem of medicine bottle sorting,the main work is as follows:(1)Combined with the characteristics of the medicine bottle,this paper designs the overall scheme of the medicine bottle detection system and builds the relevant detection system platform.(2)The binocular vision is calibrated in the medicine bottle detection system.The internal parameters and distortion coefficients of the monocular vision,the rotation and translation parameters between the left and right cameras,and the stereo correction of the binocular vision are obtained by using the chessboard.(3)The image taken by CCD camera is segmented by threshold,then the image of medicine bottle obtained is preprocessed enhanced.In view of the problems of many parameters of full connection and low recognition accuracy when VGG network trains the medicine bottle image,this paper proposes an improved VGG network algorithm,which improves the recognition accuracy of the medicine bottle image by deepening the network structure and reducing the dimension of the feature image.(4)In the process of VGG network training,the network is difficult to converge and the accuracy is not high.This paper proposes an improved network pre-training method.Through the unrelated data set training network,the parameters of the first several layers are shared to the parameters of the VGG network,and the parameters of the later layers are initialized,which accelerates the model training and convergence and improves the recognition accuracy of the model.(5)Aiming at the problem of medicine bottle sorting,this paper adopts the robot control strategy based on visual recognition.Through the improved VGG network algorithm to identify medicine bottles,then sends the position and category information of medicine bottles to the robot.According to the recognition results,the robot adopts the corresponding grabbing method to achieve the purpose of correct classification and placement.The detecting system based on the combination of machine vision recognition and robot is able to achieve 92% accuracy rate for medicine bottle sorting,and the overall inspection effect is good.