Translated Abstract
In recent years, robot technology has been widely applied to industrial production, aerospace, medical, agriculture and other fields. Robots can replace the human in high-intensity, or high-risk job. The intelligent robot, which can be aware of the environment, espetially the visual servo robot, can adjust its own control strategy according to the environmental changes. Comparing to the ordinary ones, the visual servo robots have better adaptability, which can widely expand their application fields. So, the visual servo robot has been a research focus for many years.In this paper, the author makes a in-depth study research on two key technologies: visual servo control algorithm and target recognition algorithm of manipulator visual servo system, using a seven-degree manipulator system which is based on binocular stereo visionAchievs a image-based visual servo algorithm which can control the manipulator reach the desired position and posture to do the task without calibrating the system parameters Proposes two target recognition algorithms, used to find the target object of the servo task in different conditions.The image-based visual servo algorithm achieved in this paper, uses image Jacobian matrix and robot Jacobian matrix to establish the relationship between the image feature error and the angular velocity of the joints. The image Jacobian matrix is estimated by the method of least squares, and the robot Jacobian matrix is calculated by the differential transform method according to the manipulator model parameter. When getting the joint angular velocity, weighted least norm method is used to optimize the joint angles, which are better to be away from the limits, to get more operating space.For target recognition, an algorithm based on a mixed characteristics of LBP and RGB, using the maximum posterior probability as the index of similarity, is proposed. This algorithm has a good performance in complex background and illumination-changing environment. Another algorithm is proposed for the mechanical parts to recognize in the subject, builds a model with parameters of the target object, and calculates its contour in the image. Then, compare the contour to those ones detected in the image, using particle swarm optimization method to search the optimal similarity of the comparing contours, and thus to find the best matched image area. At the same time, the position and orientation information of the target can be obtained.Finally, the author writes the control software in VC environment to achieve all the algorithms above, and completes the visual servo task of recognizing, picking and placing the specified object.
Corresponding authors email
