Research On Button Detection System Based On Image Processing

With the rapid development of computer technology in modern society,image processing is widely used in textile,instrument manufacturing and other industries.Aiming at the problem that most button manufacturers use manual methods to implement detection of button quality and counting,which leads to low detection efficiency.And a button defect detection and counting system based on image processing is designed,which mainly consists of four parts,i.e.button detection and positioning,defect detection of the plastic button with holes,defect detection of the metal button with text patterns and button counting.The main purpose of button detection and positioning is to locate and extract ROI.Two methods for button detection are proposed.The first method,based on template matching,applies image pyramid and OSTU algorithm to to-be-inspected and template images.Then the template matching is used for the binary image to complete the button detection and ROI extraction.The second method,based on Hu invariant moments,in which median filter is used to denoise firstly,OSTU segments images according to threshold,and then contour extraction is implemented.What’s more,it removes contours with noise and detects contours of circular buttons by Hu invariant moments.Finally,The Hu moment invariant matching method with higher real-time performance is selected.The main purpose of defect detection of the plastic button with holes is to find out the deformation of inner holes,loss of inner holes,edge damages,stains,scratches and pits.A defect detection method based on contour is designed.Firstly,extract button regions and fetch related information by OSTU threshold segmentation and connected region marking.Secondly,judge whether inner hole missing and color defect exist according to the number of contours.Eventually,circumscribed circle variance contour defect detection method is proposed,as a multi-step contour detection algorithm,where minimum circumscribed circle is used to detect the roundness of button contours and circular contours are transformed into the curve contours.Next,the global smoothness is detected and verified based on the cubic variance.In the end,the local maximum quadratic variance is applied to detect the notch.The experiment results show that the designed algorithm has a remarkable improvement on the accuracy of contour defect detection and meet the requirements of real-time response in comparison with the minimum circumscribed circle roundness evaluation method.The main purpose of defect detection of metal buttons with text patterns is to position the defects such as scratches,grooves and wrong text patterns.Thus,ORB+KNN+PROSAC for this kind of defect detection is proposed.At first,ORB algorithm obtains feature information of buttons,with o-FAST extracting feature points of button images and r-BRIEF extracting corresponding feature descriptors respectively.Then KNN goes for feature matching,and PROSAC with better real-time performance is selected to refine features,which finishes the image registration according to the transformation matrix.In such way,image differences are used to realize the detection of defective button blocks in the end.As for the experiment,five common feature detection algorithms are chosen to extract button features and compared by their performance.Also,a contrast experiment is carried out on two feature matching algorithms,namely BF and KNN,and shows their effect and performance in terms of defect detection.Finally,the result indicates that the defect detection based on ORB+KNN+PROSAC has great advantage on detection efficiency and real-time effectiveness.The main purpose of button counting is to implement the on-line real-time counting of multi-line buttons on conveyor belt.A multi-line button counting algorithm based on YOLOv3 is put forward.First of all,the rotation of button images serves for data augmentation,and trained YOLOv3 detects and positions buttons,getting the center coordinate of the detected buttons.And then a vertical rectangular box channel is placed in the center of the picture for button counting.Meanwhile,any button framed by the box would be set up a horizontal channel.The idea based on the software chattering elimination,reduces the counting error caused by position offset,which combines vertical and horizontal channels to count multi-channel buttons.The experiment result shows that the designed counting algorithm can not only count buttons effectively,but also meet the response requirements of online detection.