Moving Online Bar Counting Based On Target Recognition

Moving online bar real-time counting system is a determine process in production and marketing for steel mill. So far, Manual counting is the main way for steel bars. This may result to counting error because of visual fatigue. At the same time, this low efficience counting method can not catch the development of high automation requirments which will affect the reputation of the enterprise and economic benefits. In order to resolve above mentioned problems. This paper aims at developing moving online bar real-time counting system for track and counting sequence images of moving bar. Three parts are needed in this system: CCD camera is used to collect sequence images in steel bar production line; bar counting for static image and using target tracking to relize real-time counting for sequence image.Firstly, pretreating the image, then adopting a median filter method which is fit for ending preparation to smooth the image through compared the median filtering method with the mean filtering method. This fliter arithmetic is chosen to reduce the impact of noise on image and to protect the image edge information. The main methods such as the iterative threshold method are using for image segmentation, and then comparing and chosing otsu method to produce necessary binary images.In the process of bar recognition, this paper chooses the improved edge detection, center gathering, clustering recognition and other methods to dispose single-frame image. Sobel edge detection is used to detect edge information(edge intensity, gradient direction and caculating based on the absolute value equal principle) in order to extract bar edge information and reduce the time that the caculates take. Edge-center cluster arithmetic is used to enhance the target center. Then, the center of the target area is primarily confirmed by aggregation, and the number of the bar in single-frame images is recognized. Since the focus of this paper is moving online bar real-time counting system, tracking and contrasting are needed for sequential frame images to realize moving bar counting in chain grate. For all links, this system chooses target tracking algorithm based on feature to recognize counted and counting bars by estimating horizontal displacement between two sequential pictures. This paper uses fault tolerance algorithm to avoid counting unnecessary cluster regions and leaving out steel bars which should counted, setting appropriate confidence value to ensure the accuracy of counting.The hardware design must give attention to extreme production conditions and special requirments of image processing. The hardware architecture consists of Balser, IVC, IPC and lighting system. This design uses Visual Studio 2010 to develop a quick and convenient applacation and a convenient interface based on MFC.The experiment results show that the precision of that designed moving online bar counting system can reach 93.3%. And the processing time of single-frame image can satisfied the demand of moving online bar counting system, which has a very practical value and favorable market prospects.