Research On Bundled Steel Bar Face Recongition And Counting Method Based On Image Analysis

Steel bar is the necessary raw material for building and all kinds of facilities.With the growing demand of building industry,the counting problem of bundled steel bars is very important in checking.The existing methods for automatic counting of steel bar include photoelectric method,RFID label method and image recognition method.Photoelectric method has a large result error in counting bundled steel bars,which is not reliable.RFID label method has high accuracy,but it has not been widely used because of high cost and high standard environment.While the image recognition and counting method has low cost,high accuracy and most of it's work is completed by the computer,it has become a hot issue to solve the problem of steel counting.The original image of steel bar has many blemishes,such as clutter,low contrast,and much noise.What's more,steel bars are stacked tightly and the face shape is irregular.All of these make it difficult to recognize and count the steel bars.In the image analysis and recognition,the preprocessing and binarization of the image will affect the accuracy of counting results.Aiming at counting the bundled steel bars in the image,we will study and improve the image contrast enhancement method and the binarization method,and propose a new method for recognizing and counting steel bars.The main research work and innovation are divided into three following parts.(1)Optimize the gain coefficient and the enhancement way of the local Adaptive Contrast Enhancement method.Aiming at the problems of complex background,insufficient contrast and uneven brightness of steel bar images,we select local Adaptive Contrast Enhancement algorithm to do further research and improvement.Due to the limitations of the local ACE algorithm,we use standard deviation as the gain coefficient for improvement,and selectively enhance the image through comparing mean and standard deviation of local and global.The experimental results show that the improved local ACE algorithm can significantly enhance the image contrast and highlight the target details in the image.(2)Optimize the threshold function and selection range in the OTSU segmentation algorithm.By studying and analyzing commonly used steel bar image segmentation methods,combined with the characteristics of reinforcement images,we choose OTSU method and design an improved method.Firstly,according to the class cohesion are ignored in OTSU,we introduce the intra class variance into OTSU threshold function,which makes the image segmentation threshold closer to the actual one;Secondly,according to the research results of the threshold criteria,we weight the target variance of inter class variance to adapt diverse target ratio;Thirdly,considering about the characteristics of steel bar image,we reduce the selection range of threshold,and the amount of independent calculation is reduced in this way.Finally,combine the three points and the OTSU is proposed,and the experimental results show that the threshold selected by optimized method has a better segmentation effect.(3)Propose a classification matching recognition and counting method based on multi features of connected regions.Based on area counting method and template matching method,and the characteristics that most of steel bar faces are single and have smaller proportion of bonded reinforcement in the binary image,a classification matching recognition and counting method of reinforcement based on multi feature fusion of connected regions is proposed.First,the target area is divided into four types according to the area characteristics.Second,different features are used to identify the target areas of each type.Among them,we judge the number of the incomplete separation based on Euclidean distance of their barycenter.The single end face is recognized by area simply and fast,which identify most of the steel with a small amount of calculation.For the bonded reinforcement,the area,shape factor and barycenter feature are fused to recognized targets and determine the number of bars which eliminate the bad influence of end adhesion.The noise do not participate in counting.The results of comparison experiment show that the classification recognition counting method proposed in this paper has a high accuracy rate of steel counting.