Dispenser Automatic Positioning System Based On Machine Learning And Machine Vision

With the development of electronic industry,manufacturers have designed thinner and lighter products,which need a more sophisticated packaging process.Traditional dispensers have a series of problems,such as poor precision and low efficiency.Therefore,this paper adds machine vision technology and machine learning algorithm to the traditional dispensers,which make the dispensers have a higher precision and efficiency so as to meet the needs of modern manufacturing.This paper mainly introduces the configuration of a machine vision based disperser system.The system mainly consists of two parts: hardware and software.The hardware includes light sources,cameras,lenses,and motion controllers,and this dissertation analyzes their roles in machine vision system,and discusses the performance parameters and model selection.The software mainly includes modules such as image preprocessing,camera calibration and object detection.The image preprocessing converts the three-channel color image into a single-channel gray image,to reduce the amount of image data and improve the computational efficiency,and then suppress the disturbances from external environment by image filtering.The conversion relationship of four coordinate system is analyzed by means of a typical pinhole camera model.The lens distortion existing in the actual imaging process is also considered and corrected by means of the functions of OpenCV in visual image processing modules.According to the actual situation of visual dispenser,a system calibration method suitable for visual dispenser is selected.Afterwards,according to the actual situation of the machine vision based dispenser,this dissertation uses two object positioning methods:(1)combination of the MB-LBP cascade classifier with two additional classifier,SIFT feature-based classifier and SURF feature-based classifier.By this way one can form a new cascaded classifier meeting the requirement of object inspection and positioning of the dispenser.(2)The HOG and SVM based object detection and positioning.Because of the large dimensions of the HOG feature,PCA is used for dimensionality reduction and then the SVM is used for classification and positioning.The two methods achieve a similarly good performance and both of them have been applied successful in visual dispenser system.