Development Of Thin-Sheets Automatic Counter Based On Image Recognition

The counting of sheets occupies an important position in the trading process of industrial products,and in the sheet product industry and related fields,each company is based on the number counting of sheets to carry on relevant business activities.Therefore,in various processes during business,the most crucial step for company is to count the number of sheets,while ensuring the counting speed and accuracy.The current single sheet thickness in the market is between 0.6~2.0mm,the traditional counting methods such as manual counting,thickness measurement or weight measurement exist errors and appear low efficiency,seriously affect the development of enterprises.At present,the Tianjin Chen handheld CY-6800 A board device on the market is the first instrument to count the production of printed circuit board(PCB)in China,the use of laser counting,but there is no instrument that is combined with computer vision,in order to meet market demand,aiming at problems that the sheets are not absolutely parallel and slightly deformed,a method of counting pixel jumps based on skeleton extraction is proposed in this thesis.This thesis contains four modules,respectively are image acquisition,object extraction,texture extraction,multi-evidence counting based on gradient direction.The target images are captured by mobile devices,such as mobile phone and tablet.Aiming at the characteristics of image texture,Gabor filter is used to enhance the target edge based on a specific direction and sobel algorithm is used to extract the edge.Mean Shift algorithm is used and then the grayscale image is binarized,which effectively reduces the interference point.In order to remove the interference left and provide clear targets for subsequent counting,this thesis uses water-filled algorithm,morphology and BlobsFiltering method to eliminate the residual interference in images.The obtained binarized image is subjected to single-pixel skeleton extraction.Finally,according to the two innovations proposed in this paper,the starting points and the end points of the counting are automatically obtained by using the mathematical features of the texture respectively.And according to the projection in the gradient direction the to automatically obtain the target area and the probability statistics,effectively overcomes the problems of small contact gap between the thin plates and slight bending which is difficult to identify.In practical application,the principle of perspective transformation for automatic calibration is combined with four points recognition,the algorithm can accurately realize automatic counting of sheets in side photos.It can effectively replace the existing sheets measurement method,which can reduce the factory's labor and material resources consumption and help the factory break through bottleneck of development.Experiment results show that the images captured using a 800-megapixel tablet camera canbe used to count 285 sheets with 1.2mm thickness,and the average sheet repetition period is 11 pixels.The pixel hopping counting algorithm based on probability statistics effectively solves the problem of counting non-parallel sheets,the precision of the algorithm can meet the requirements of enterprises production.