Vision Based Serial Manipulator Kinematics Identification

The goal visual based kinematic identification is to estimate kinematics relationships from given image data. In this work we propose a solution for vision-based kinematics that estimates Transformation matrices from given set of images of robot instrument with visual marker mounted on it. This problem is one that attracts a lot of interest from robotics community, such solutions provide robotics systems ability to use control system solutions for wide range of robots and robotics structures and provides ability of rapid robot systems development due to ability of robot to estimate it’s kinematics autonomously. With the rise in use of computer vision systems and especially visual control solutions of robots, many researchers have turned to computer vision-based robotics control practices. In this work we address a algorithm solution to estimate kinematics relations in terms of transformation matrices, objects kinematics description directly from image data of end-effector motion. We propose a solution to determine transformation matrices of each link by sequentially moving links and recording images sequences of motion. Proposed algorithm allows to estimate transformation matrices even with noised data.