Abstract: In order to calibrate the selfdeveloped industrial robot and improve its movement accuracy, a robot calibration method of dynamic tracking spherical target number by binocular vision was put forward. First we carried out the arbitrary position and orientation measurements in the workspace by markedtargets set on the sphere surface installed on the robotend, which was distinguished and numbered by the particular arrangement and distribution, then accurately identified the geometry parameters of the robot by a leastsquare iterative and compensated them to controller.We established the calibration software including vision measurement, data processing, robot control and other modules by using MFC based on open, modular ideology, finally verified its reliability and accuracy by measuring examples and comparative experiments. Experimental results show that the pose data measured by the software is precise, expanding the field of view of traditional visual tracking; and the identified actual geometry parameters can be compensated, successfully improving the robot positioning accuracy from 3.785 mm to 1.618 mm, as well as the orientation accuracy from 0.235 to 0.139.