Research And Implementation Of Large Field Alignment System Based On Binocular Calibration

Machine vision has been widely used in the transformation and upgrading of traditional industries.With the rapid development of China's economy,the gradual increase of labor costs and the strong support of national policies,the research on machine vision gradually becomes significant.At the same time,there are some very dangerous situations in industrial production,and some of them are impossible to do,or because of the error of manual operations,the industry is increasingly using the machine's visual skills to conduct product quality testing,defect detection,counterpoint,and motion tracking.Based on the significance of this research,the paper studies the counterpoint of the large-field scene in the packaging industry by using the machine vision technology,and develops a complete set of counterpoint system to cooperate with the robot manipulator.The main work of the paper is summarized as follows:(1)For large industrial products,the view of a camera cannot cover the key parts of the target,and a binocular vision calibration method is proposed,and a specific calibration plate based on the coding circle is designed for this method.Several experiments were designed to verify the accuracy of the experiment and meet the actual production needs.Aiming at large industrial products,a camera's field of vision cannot completely cover the whole object,and a binocular vision calibration method is proposed,and a specific calibration plate based on the coding circle is designed for this method.(2)It elaborates the hardware selection and software composition of the system in the whole wide field of vision,and introduces the whole system framework with the actual software interface and field application pictures.Large number of experiments have been design,that accuracy of calibration,precision,stability and time time of alignment of the whole system are analyzed,and finally,the accuracy of the system is high,stability is good,time consuming and suitable for production practice.(3)A high robust line fitting algorithm based on supervised learning is proposed.The ordinary least square line fitting method is greatly influenced by the outliers and increases with the increase of the number of samples.In this paper,the weight coefficient is added,and the sampling is carried out,which can reduce the time consumption,and the mechanism of supervision learning is added to improve the fitting accuracy.(4)In the discussion of the RANSAC circle fitting method based on the standard round hough transform,the concept of relative roundness is introduced to determine whether the remaining point is the local point,so that the threshold can adapt to the rotation and scaling of the circle.In the analysis of the properties of the connected domain,the definition of the roundness is defined by the variance of the distance from the point to the center of the region.In addition,the method of sampling based on Angle and removing extreme value is added.This method is used to screen connected domains,which can be more anti-interference.These efforts result in more accurate central extraction on the calibration board and improve overall system accuracy and robustness.