| Machine vision has its outstanding timeliness,stability,and reliability,and is widely used in many fields such as workpiece defect detection,dose detection,chip pin detection,and express sorting.In the field of large-scale box sorting,the manipulator that equipped with a machine vision system is equivalent to giving the manipulator an "eye",which is promoting the production and operation of enterprises to be more intelligent and flexible.Machine vision guiding manipulator position have two main methods: Eye-in-hand and Eye-to-hand.Eye-in-hand means that the camera is fixed on the manipulator and moves together.This installation method can observe the camera perspective at any time.The background of the image changes when the angle of view is switched,which brings the complexity of image segmentation in later.Eye-to-hand means that the camera is fixed at a certain height and does not move with the manipulator.Although there are certain restrictions on the flexibility of the manipulator,the background of the captured image is almost stationary and easy to segment the image.With the camera installed in Eye-to-hand mode realizing the guidance position of five-degree-of-freedom manipulator carrying with monocular vision.I learned about relating theories about machine vision and robotics for project preparing in the early,establishing research routes at the same time.Furthermore,the research on machine vision system calibration,manipulator motion control,target recognition and positioning algorithms is carried out.The main research contents include:1.First,establishing the conversion relationship between camera calibration and hand-eye calibration model of machine vision system.In the camera calibration process,Zhang's checkerboard calibration method was used to obtain camera internal parameters.Then theoretically analyze the relationship between the camera status and the end-tool status conversion,and using Navy hand-eye calibration algorithm to solve the hand-eye calibration conversion matrix on Matlab.Secondly,establishing D-H model to perform forward kinematics analysis,according the forward kinematic conversion matrix to figure the inverse kinematics parameters,which verified the non-uniqueness of the inverse kinematics solution.Finally,the point-to-point single-segment trajectory planning was implemented,researching the effects of cubic and quintic polynomial interpolation on the motion stability of themanipulator.The experimental results confirm that the angular displacement and angular velocity curves in the joint cubic polynomial interpolation simulation are smoothly transitioned,which meets the requirements of manipulator motion stability.2.In the aspect of target recognition,firstly,the mixed color filtering processing is performed on the collected color image after gray value conversion.The filtered image saves better image edge information.Because high quality image has its obviously different target level between target and background,so in order to obtain better image quality,image enhancement can be used to make the target boundary clearer.Then the threshold method(maximum inter-class variance method)is used to achieve the segmentation of the background and target.The segmented target has a better edge continuity by morphological closed operation.Finally,the strongest operator Canny algorithm is used to get the edge information of the target.The extracted target pixels are almost sub-pixels,preparing for subsequent target positioning.3.The "one-eye positioning method" is used in image target positioning,where assisting with the help of artificial marks to realize rapid positioning in single frame image.Using the improved Harris corner detection algorithm to detect the corners for estimation of target centroid coordinates in two-dimensional plane.Then,the depth information of the target is solved by using the P4 P point feature for the artificial mark pattern attached to the target surface.We have completed the position matrix of the target in the camera coordinate system so far.In addition,the target rotation matrix relative to camera status can be solved by using three artificial landmarks that are not collinear.Because the camera is fixedly installed,it is equivalent to the target object rotating relative to the camera.So far,the Euler angle of the target object can be obtained.Combining the above-mentioned position matrix with the Euler angle of the target,the pose of the target under camera coordinates is solved.The experimental results show that the proposed combination of corner points and artificially assisted identification points to achieve fast and accurate positioning of the target pose.The experimental results prove that the average relative error of the position in the Z-axis direction under the world coordinate system reference is 7.096% The average relative error of the yaw angle and attitude estimation is 19.58%.The above data proves that it has a certain reference value in achieving rapid target positioning. |
PDF Full Text Request