Research On Key Technologies Of Automatic Alignment System Based On Visual Feedback

In recent years,with the development of high-speed data processors,alignment systems have become more and more widely used in the hoisting and handling of large objects,such as container hoisting,material loading and ammunition transportation.By measuring the spatial position of the target,the moving parts are automatically adjusted for alignment.In this way,it overcomes the inefficiency of the traditional semi-automatic alignment method using manual observation and manipulation.Furthermore,the system can complete the alignment task well even when it is very close to the alignment target,the working space is small and the surrounding is blocked by other equipment.In order to improve the automation and intelligence level of the alignment system,this thesis studies some key technologies in the missile hoisting and alignment system.The technologies include measurement and path planning.The main research work and results of the thesis are as follows:Firstly,the related process and framework of the automatic alignment system based on visual feedback in this thesis are described,and the mathematical representation methods and models related to visual measurement and alignment system are introduced,including the visual measurement model,the D-H parameter representation method of robotic arm kinematics and the forward and inverse kinematics models of the robot arm based on the homogeneous coordinate transformation matrix.Secondly,since each measurement camera corresponds to a cooperative target point,there is no common target in the alignment measurement system.Aiming at this situation,an overall vision measurement solution is established.An on-line adjustment and calibration scheme of the measurement system without changing the system structure is proposed.By using the online scheme,some tasks can be completed,including the adjustment of the verticality of the optical axis of the optical measuring device,the calibration of the focal length of the measurement camera,the calibration of the laser ranging data,the calibration of the pose relationship between the two cameras,and the calibration of the hand-eye relationship of the measurement system.Focusing on solving the problem of calibrating the pose relationship between two non-common target cameras in the measurement system,a binocular camera online calibration method is proposed based on the DLT-LM algorithm.The experimental results show that the point distance mean square error obtained by the binocular camera calibration method based on the DLT-LM algorithm is better than 0.2mm.Then,a measurement method is introduced to realize the pose measurement of binocular camera without common target point.The problem of measuring the horizontal alignment deviation of components in the case of only two cooperative target points is analyzed.By establishing the coordinate system of the measurement system,the relationship between the imaging equations of the two cooperative target points is constructed.Furthermore,the horizontal alignment deviation between the upper and lower parts can be obtained by combining with the biaxial inclinometer and the calibrated measurement system parameters.An error analysis model of the measurement system based on the Monte Carlo method is established.The horizontal alignment measurement mean square error between the upper and lower parts of the alignment in simulations and experiments.The results show the mean square error does not exceed 1mm,which meets the actual demand of missile hoisting alignment.Finally,the robot motion path planning and obstacle avoidance methods based on Dynamic Movement Primitives(DMP)are studied.In the visual feedback system based on position information,the DMP method can re-plan the execution path according to the position feedback information,therefore,it can complete the alignment task even when the position of the object to be aligned changes.In order to enable the robot arm to automatically avoid possible moving obstacles during the alignment process,Kalman filtering and model predictive control algorithms are used in the DMP framework.Kalman filtering is adopted to estimate the relative state between the end-effector of the robot arm and the obstacle in the future.Model predictive control is employed to optimize the robot action to achieve a better obstacle avoidance performance.