Real-time Recognition System Of Steel Bar Based On Machine Vision

In order to ensure the standardization of packaging bar, the online counting becomes amajor process on the production line. Currently, the bar counting is mainly completed bymanual labour, this method can make the workers get fatigue easily, especially prone to errorcount, which can not guarantee the number of packaging requirements and match the highlevel of automation equipment of steel rolling production line. To solve these problems, theautomatic counting system of bar is developed in this paper, which firstly identifies thenumber of bars to the single-frame images and secondly achieves real-time counting ofmoving images. The identifying process comprises four components: the acquisition ofcontinuous images by a CCD camera, to filter and segmentation the image, the bar countingin a single image, tracking the moving target to realize the real-time counting of images withcontinuous bars.The image preprocessing includes image filtering and segmentation: This paper selectsthe median filtering algorithm after comparing advantages and disadvantages of the mean andmedian filtering method, which can both eliminates the image noise and protect theinformation of edges; A variety of methods and a new promotion of OTSU algorithm are usedto separate the image in this paper, this system selects the promotion method for imagethresholding which has a good segmentation result.In this paper, a new method is promoted to identify bars of single image, it concludesedge detection, center gathering and cluster identification. Firstly, adding the templates ofSobel edge detection operator from two to eight, so it can get the edge information of eightdirections, which includes the edge strength and gradient directions. When do the calculation,it can only convoluting the templates in four directions based on the principle of equalabsolute value, which can extract the edge information effectively and reduce the amount ofcalculation greatly. Then using the direction information to gather the edges toward the center.Nextly identifying the center of the position of the target area by clustering methods. Finally,making a judgement of the candidate target clumps features, removing noise and any otherfactors that caused by interference information, thereby we identify the number of bars in thesingle image. Since collected the continuous rod images by camera, Therefore, to achieve the track counting of continuous images, the bar needs to be tracked and aligned between theseries images. So we select target tracking algorithm based on bars’ feature, which canobtaines horizontal displacement of corresponding bar in two successive images. Then made adetermine of which bar is counted in the image. At last, a fault-tolerant algorithm is adopted,only when the number of matches reaches the set value, the bar is counted eventually.The hardware system of this system is composed by3parts: Balser industrial cameras,industrial machines as well as blue and white LED light source, they are able to adapt theharsh production environment in the production site and meet the requirements of imageprocessing algorithms. In the software part, Visual Studio2010is used to develop applications,the Microsoft Foundation Class Library MFC is used to develop a simple and convenient userinterface.In this paper the effectiveness of the adopted image processing algorithms are verified byexperiments, the results show that the accuracy of bar counting in a single frame image is upto96%for this system, and its processing speed can meet the online real-time requirements, italso has a good market prospects and a high practical value.