Research On Target Positioning Based On Machine Vision

With the continuous development of artificial intelligence technology,robots are gradually infiltrating into every aspect of human life.In assembly lines,the execution action of the robot is preset by off-line programming.However,when the working environment changes,the robot does not have the ability to cope with the change.Therefore,the vision technology is built in the robot to obtain the vector information of the target in real time,in order to make the execution of the robot more flexible.Taking the real-time acquisition of target position vector information by vision sensor as the application background,this thesis studies the location of static target and moving target by using monocular vision technology.The specific research contents are as follow:Positioning Location analysis of static target.First of all,the industrial camera is calibrated.The traditional camera calibration method is directly carried out in the camera calibration toolbox in MATLAB.In this paper,based on the C++ development language,the camera calibration program is written to get the internal and external parameters of the camera.The average calibration error is 0.2 pixel,which is 0.5 pixel lower than the general calibration error.Then the image captured by camera is preprocessed.In the image edge detection,the traditional Canny edge detection adopts the form of Gauss filter,which may cause the image edge to be too smooth,resulting in the loss of image edge details,the reduction of edge extraction ability and other problems to be optimized.The form of bilateral filter is proposed to replace Canny edge detection The Gauss filter in the measurement completes the image edge detection processing.Then,an efficient method to obtain the target vector information is proposed for the preprocessed image.The method of finding the principal axis is used to find the smallest external rectangle including the target area,and the centroid coordinate value and deflection angle of the target are determined.At the same time,the micro adjustment amount is set in the actual operation process to reduce the error.Finally,based on the C++ development language,the human-computer interaction interface is designed in QT creator,and a series of experiments are completed.The experimental results show that the average error of static target positioning is 0.95 mm,which is lower than the2 mm error of general visual positioning system.It has been applied in the actual productionline.Location and tracking analysis of dynamic target.Compared with the traditional methods,that is,directly preprocessing the captured image,this thesis proposes to use the Grab cut method to extract the foreground of the captured image to reduce the influence of the background on the target location,and then preprocess the extracted foreground.In this thesis,an efficient method to obtain target vector information is proposed for the preprocessed image.The method of finding the principal axis is used to determine the minimum circumscribed rectangle containing the target,and the centroid coordinate value and deflection angle of the target are determined.Finally,the design of human-computer interface is completed in QT creator based on C++ development language,and a series of experiments are completed.The experimental results show that the average positioning error is 0.95 mm.Errors below 2 mm General Visual Positioning System.Finally,the location of the moving target is analyzed in this thesis.The GMM Gaussian mixture model is used to eliminate the background of the real-time video stream,and the limitation of target tracking and location based on color features is discussed,and the camshift algorithm is used to chase on the basis of Meanshift algorithm.The saturation component and kalman filter are introduced to optimize the problem that the target is occluded in the process of moving,which leads to the failure of camshift tracking and location.The Babbitt distance between the target area and the calibration area is calculated to determine whether the target is obscured or not.It is proposed to set the minimum region area for the occluded target,and by comparing the identified region area with the minimum region area,the filtering parameters are updated and the position information of the next frame of the target is predicted.The experiment is carried out in VS2017,and the experimental results show that the optimized algorithm can also predict the position of the target in the next frame after the target is occluded,obtain the centroid position of the target,and achieve the positioning effect of real-time tracking.