Research On Robot Control With Position-based Visual Servoing And Target Recognition Methods

Robotic vision technique is an important research field on the machine vision, and it's a new subject based on the development of the computer vision theories and the image processing techniques. Nowadays, much attention had been paid to it by the researchers on robotics, and robotic vision techniques have been successfully applied in many fields. In this paper, based on the GBR-400 vision robot systems with 4 degree of freedom (DOF), the recognition and grab on the single target object and multi-target objects are studied by using the method with position-based visual servoing, and some findings and conclusions have been achieved.Through analyzing to GBR-400 vision robot systems in detail, the basic structures of the position-based visual servoing control systems which include cameras and its calibration, digital images capturing and processing, coordinates transform, modeling and analyzing of robotic manipulator kinematics, path planning and servoing control system, are built. The built structures are dependable foundation for the following research in this paper.Based on the selected target objects, relevant image preprocessing techniques and algorithms which include image filtering, image enhancement, edge detection and image segmentation were analysed and compared, and appropriate image preprocessing algorithm were chosen for the visual servoing systems discussed. The kinematical model of GBR-400 robotic manipulator with 4 DOF was established by using D-H method, and the kinematical inverse solvings were also given out, which are theoretical basis to robot path planning; The camera calibration techniques were studied, a simple method with applied linear calibration was proposed, in which the nonlinear model was transferred into linear model. The calibration reduces the requirements for experimental conditions and simplifies the camera calibration process. It is applicable that the calibration method was verified through experimental study. The binocular vision robot systems with position-based visual servoing were studied, and the system structure, technique related to digital image processing, the recognition of single target object, analyzing of kinematical model and robot path planning as well as visual servoing control were analyzed and studied. After analyzing the traditional image template match algorithm, a recognization method with total pixel-based matching was presented, and experimental results shown that the presented method is validity. For the multi-target recognition of robot visual servoing systems, a method based on improved Hu's invariant moment and Euclidean Distance as the measure of target similarity was put forward. The experiments about recognizing, selecting and gripping for five different target object proved that the proposed method is effective. At the same time, it was shown that it is feasible that the invariant moment is used as feature to recognize multi- target objects. Further, it is seen that robot visual servoing based on the features will be a direction with quite development potentiality for the future robot visual servoing control.