Monocular Vision Location System Of Polishing Robot With Four Degrees Of Freedom Based On The Distance And Plane Constraints On Feature Points

There is a new method which automatic polishing large-scale free-form surface by micro-robot. In order to enable the polishing robot moving along set route, robot should be able to determine their location and movement by self. In related technology of Autonomous Mobile Robot Research, the navigation of mobile robot is not only a key technology, the base of the implementation of the robot mission. The goal of navigation research is the mobile robot is able to have a destination, finish an assigned task and carry out a specific operation. Localization for mobile robot is the most basic part of navigation system and the most basic problem which need be solved for achieve tasks of navigation. Real-time and precise localization is the key to improve the performance of robot and the guarantee to realize the robot autonomous navigation. In paper, we established the localization system of polishing robot with four degrees of freedom based on monolar vision according to the actual needs of the free-form surface polishing microrobot and advantages and disadvantages of different positioning methods. The localization system is able to provide the parameters of the position and direction of robot for control system of polishing microrobot.Localization system of polishing robot is mainly made up of seven parts, high-resolution CCD camera, fixed focal lens, optical filter, image acquisition card, computer workstations, optical feature points arranged in the robot body (infra-red light-emitting diode) and positioning software package. In the system, the six infra-red light-emitting diodes as measured three-dimensional information carriers and their coordinates at the robot coordinate system have been accurate calibrated by coordinate measuring machine. The positioning point's coordinates in the robot coordinate are known. The images of feature points on the robot are taken by CCD and the visible light is flitted by 760mm wavelength optical filter. So the six infra-red light-emitting diodes is imaging in photosensitive area through the fixed focal lens. Then the light signal is turned into the electrical signal and transmitted to the signal processing circuit. In the signal processing circuit, the electrical signal generated by photosensitive component is amplified by amplifier, completed A/D conversion and outputted in the form of digital video. Image data is transmitted to the image acquisition card through CameraLink 1.10 data interface and turned into a standard 8-bit grayscale images. After image processing, them is transmitted to the computer for further treatment through high-speed PCI-X interface. In the positioning software package, the centers of feature points are calculated and we can obtain their coordinates at pixel coordinate system. Make use of the feature projection points'coordinates and feature points'coordinates at camera coordinate system, combined with the constraints of the location between six feature points in robot coordinate system, the conversion matrix between the robot coordinate system and the camera coordinate system will be obtained. Then the coordinates of target point at camera coordinate system can be calculated. Use of the conversion matrix between the camera coordinate system and the world coordinate system which be obtained from camera calibration to calculate the target point's three-dimensional coordinates at the world coordinate system. Therefore, the current location parameters, the Attitude Parameters, as well as coordinate transformation matrix of the polishing robot is obtained and sent to the navigation, control and movement programs to control the robot to operate the next exercise.According to the above-mentioned principle, the localization system of the polishing robot is established based on the central projection model. The process of model building and algorithm are be introduced and explained in detail in the paper. In accordance with the disadvantage of monocular vision, the positioning accuracy and stability at the depth direction is less than at other directions, we proposed an error correcting technology to correct feature points'coordinates at camera coordinate system by means of fitting plane. First of all, we imposed the planar constraints on the layout of feature points on the robot in order to make six feature points located on the same plane. Secondly, we fitted a plane by feature points'coordinates at the camera coordinate system which calculated by programmer. In order to improve the accuracy of fitting plane, weighted fitting method is utilized to obtain fitting plane according to the accuracy of the calculation of different feature points. The inverse terms of distance between feature and the equal-weight fitting plane as feature points'fitting weight. Finally, we made vector between feature projection points and the optical center of camera as direction vector of a straight line, then the feature points'coordinates were replaced by the intersection points'coordinates of straight line and fitting plane. The method of correcting feature points'coordinates is able to improve the accuracy and stability of feature points'coordinates at camera coordinate system by the experiment.The calibration of camera is the important step in computer vision, but also a necessary condition for acquires the accuracy information. Camera calibration is process of establish a camera imaging model and aculeate parameters in it according to image information and relation between points in three-dimensional space and their projection in two-dimensional imaging. In the localization system of polishing robot, Zhang zhengyou's plane template two-step method has been used for camera calibration. The technique only requires the camera observe a planar pattern shown at a few different orientations. Either the camera or the planar pattern can be freely moved. Compared with classical techniques which use expensive equipment such as two or three orthogonal planes, the Zhang zhengyou's plane template method is flexible, easy to use and low cost.Finally, the experimental results of were provided. The experiments include of fitting plane correcting feature points'coordinate, location precision of a single point, repeatability precision of a single point and tracking of robot experiment. The experimental results show that: The vision location system of polishing robot able to meet the requirements of real-time location of the robot.