| When thin sheet products(such as paper,cigarette package,PCB board,etc.)are manufactured and delivered to the customer,it is necessary to count the number of products fast and accurately.If the counting result is not accurate and timely,it will have a direct impact on economic efficiency of related companies.In view of the above-mentioned industrial demand,this paper uses computer vision technology to develop a low-cost counting instrument.The current vision technology of detecting stacked sheets is mainly faces with three major problems:(1)Constrained by the hardware cost,how to meet the need of detecting the thinner sheets under current fixed imaging resolution;(2)Data anomalies and deformation(rugged edge,varying thickness,uneven illumination,irregular arrangement and so on);(3)The factors of constraint between image field and resolution,unobvious features of detected image and so on make stitching images more difficult.To solve above problems,this paper presents a new imaging scheme and counting algorithm from the aspects of imaging system design,line detection and homogeneous image stitching.The main research contents are as follows:(1)From the perspective of computer vision technology,we analyzed the imaging characteristics of detected objects,and comprehensively considered the actual needs of customers and the feasibility of whole hardware structure design.Because of the problem that they cannot be improved simultaneously for single-camera field of view and imaging resolution,we propose an imaging scheme based on multi-camera staggered array.(2)Based on the statistical analysis of large data,a robust stripe detection algorithm combining global periodic constraint and local correlation measurement is proposed for the typical anomalies and deformations in the actual product image.lD profiles are extracted along the direction of stack height,and then denoised with Fourier spectrum analysis and a comb filter to preserve the useful period signal.Each candidate paper is located using a traditional peak finding algorithm.An optimal peak template is constructed,and an improved function of normalized cross-correlation is presented to calculate the correlation coefficient between the former template and the original signal by local matching.This approach help reduce false detection from complex factors such as rugged edge,varying thickness and gap,and irregular arrangement.The collinear property and similar shape of signal wave are utilized to further suppress clutter,and the ultimate measure is obtained from optimal statistics of different profile counts.(3)We proposed a fast registration algorithm for the problem in images stitching.First,the rotation angle of the image is obtained by using the fundamental phase offset information in lD profiles along the direction of stack height.And the affine transformation is used to eliminate the angle difference between two stitched images.Then,the fundamental frequency is extracted and the linear relationship between the fundamental frequencies of the images is used to eliminate the difference of the scale by the affine transformation.Finally,the corrected image is calculated the correlation coefficient,and through the Hough transform to obtain the highest matching straight line.Then the matching line is reversed into the original image,and this line can help to ultimately determine the exact position of overlapping lines.In this paper,on-line detection were carried out for different cigarette package of 0.08mm,0.1mm,0.11mm and 0.23mm thickness respectively.The statistical data shows that we can make sure the recall rate is about 99.9%under the precision rate is 100%,and there is no need to adjust the parameters for different thicknesses.At the same time,this instrument has been sold and worked at workshop.The latter product tracking survey shows that our instruments greatly reduce the input of manpower,improve production efficiency,and fully meet the industrial application requirements. |
PDF Full Text Request