Automatic Steel Bar Counting System Based On Machine Vision

The counting and packing of steel bar is an important part of steel bar production and sale in steel rolling mills. At present, the steel bar counting mainly depends on manual way in most steel factories of China. This method is low efficiency, high labor intensity and big counting error. It not only reduces the production efficiency and automation of manufacture factories, but also affects enterprise economic benefit since the enterprise could not obtain relevant economic benefit from allowed negative tolerance rolling method. So it is significant to improve counting efficiency and accuracy and realize automatic production.To meet the market requirement, an automatic steel bar counting system based on machine vision and image processing was designed. In the system, a digital camera shot the bar sections and transmitted acquired image data to the industrial computer to do image process and count. When the count reached the set value, the conveyor stop signal was sent and a separation position was indicated on image. Then the operator separated the steel bars by the separation position.Considering specific production environment and image processing requirements, the hard structure was designed and it consisted of digital camera, gigabit network card, industrial computer, lighting system and so on. The Visual C++ was used to develope a convenient and compact application interface and database SQL Sever was used to manage steel bar records. Image processing was a difficulty and emphasis in system developing. The OpenCV image processing library was called in Visual C++ to process the acquired images and realize steel bar counting. In the paper, the image processing comprised image preprocessing, target identification and target track.In the image preprocessing, median filter was used to eliminate noise and smooth the image. The Otsu method was employed to segment the image. For the limit of steel bar shearing, there may be black holes and cracks appear on binary image after image segmentation, which would lead counting error. To solve the problem, a new filling algorithm named eight directions discriminance was used to fill the holes and cracks to improve the image quality and protrude target feature.In target identification, a 16 directions edge detection operator was used to detect the bar edge exactly.16 directions edge convergence algorithm and fast template matching algorithm were used to recognize steel bars of different sizes and then Euclidean distance was used to cluster pixel points that belonged to one bar and determine the bar center. Target identification resolved the steel bar couting of single frame image. But to realized online steel bar counting, the matching and tracking of steel bars in image sequence images was essential; and this is why online counting is better than the couting of bundled bars. At last, feature-based target tracking algorithm was employed to estimate the bar horizontal displacement and track the bars.The test result shows that the precision of steel bar automatic counting system can reach 99.99% and the processing time for each frame is only 0.02s. Field use proves that the system can satisfy the requirement of auto counting and has high practical value and good market prospect.