Research And Realization Of Counting System For Steel Bars Based On Machine Vision

The real-time inspection of the quantity of steel bars is very significant during production process. Due to bad working conditions and high work intensity in production workshop, manual counting for steel bars is ineffective and has low accuracy. It is urgent to develop a high precision automatic counting device for steel bars.First and foremost, we present a hardware scheme to capture original image of steel bars. This scheme uses a uniform light source to illuminate the section of steel bars and an industrial network camera to capture the original image which will transfer to a computer in real time. In terms of light source, discuss various lighting modes, then select ring-shaped LED-arrayed light source to get high quality original images and reduce the disturbance of natural light or background in the workshop. These images are overexposure so as to improve the luminance of oxidized regions and to flatten irrelevant background.Next, pre-processing algorithms is studied in detail. To reduce the effects of uneven illumination, we transform the RGB image to gray image and do Top-hat transformation. Then we research image filtering algorithms in spatial domain, frequency domain and wavelet domain. The wavelet based filter has good performance in noise filtering. In terms of image sharing, Robert, Laplacian and LoG are used to enhance the edge information. Experiment results shows that LoG is suitable to steel bar image. To make full use of grayscale information and improve the quality of subsequent binary image,2-D threshold segmentation, Watershed and adaptive local threshold algorithm is applied to get binary image. Then Hole-filling, Convex Hull, Erosion and Open morphological operation is used to pre-process steel bar section image to further improve the binary image quality. The convex hull operation and hole filling can remove the small hole and narrow gully; Erosion and open operation can wipe narrow adhesions between the connected regions. Furthermore, we delve into the counting algorithms for steel bars. Region-growth algorithm is used to extract area, perimeter, radius and barycenter in each connected region in binary image and successive approximation method is used to get average area and average radius. Then elaborate three counting methods in detail. The fist one is Relative-Area counting algorithm. Underexposure region caused by oxidized is treated separately to do secondary image segmentation while the adhesion region caused by mutual occlusion is solved via ultra erosion and conditional dilation algorithm. Relative-Area method is taken to count steel bars. Most of thresholds and parameters are figured out according to images in this algorithm and have good adaptability。Moreover, present an adaptive template matching method based on2-D local maximum, use the average radius to construct adaptive template. Then, match the binary image with the adaptive template and get correlation coefficient matrix. Then get2-D local maximum and filter misjudgment based on Euclidean distance and display the results in original image. On the foundation of two counting methods above, we present a new combined algorithm to promote counting accuracy. The experimental results show that the combined algorithm has good performance with accuracy more than99.5%.Last but not least, application software of automatic counting system for steel bar is developed based on OpenCV and tested in actual production workshop. The software system includes image acquisition, image pre-processing, reorganization counting, manual intervention modules, display result and data storage. The online test results show that the automatic counting system for steel bar reaches accuracy of99.11%.