The Design And Research Of The Bar Tracking And Recognizing System Based On Computer Vision

China has the most bar and wire rod (hereinafter to be referred as bar) production in the world. However, its auto-counting technology falls far behind many other countries. Nowadays, most of domestic enterprises still use artificial-counting method. So the research of auto-counting system is extremely urgent. At present, most of related researches focus on bar staying in static or moving in low speed, and only counting the bar, not involve the bar quality test. Bar auto-counting is still a problem. In this article, some designs and researches are carried out to recognize and track bars:(1) According to the actual production condition in an enterprise, a set of high speed imaging system is designed and developed based on line scan camera. The components of hardware and software are introduced, and the workflow of the system is elaborated. A parameter called "undistorted acquiring parameter" and its computing method are given.(2)2-D histogram entropy image segmentation algorithm is introduced, and it is improved in this article to segment bar images. Then median filter is used to remove the noise.(3) Blob analysis method is used to extract bars’feature, and EPI analysis method is used to measure bars’length difference.(4) Greedy exchange method and Kalman filtering method are separately used to tracking bars, their workflow also are given in this article.The experiment results can bring in conclusions as follows:(1) The system in this article can continuously acquire images of bar moving in high speed. The acquired images are clean and undistorted.(2) The improved algorithm can segment images more accurately than Ostu method and2-D histogram entropy method, and also perform well on low contrast images.(3) The accuracy rate of Blob analysis result is nearly100%, and algorithm costs less than5ms; EPI analysis measurement error is less than millimetre-sized.(4) The tracking accuracy rate using Greedy exchange method is nearly100%, and the tracking error using Kalman filter is less than4pixels.In sum, bar tracking and recognizing system proposed in this article not only can recognize and track bar target, but also can measure bar’s length deffererce. And improved image segmentation algorithm also can applied in other fields.