When the manipulator stays underwater for a long time, there is a risk of accuracy degradation due to high pressure, overload and collision in the deep sea. In order to maintain the accuracy of the manipulator operation, underwater in-situ calibration is required. Since conventional precision calibration equipment such as laser tracker cannot be used underwater, this paper proposes a precision calibration method for underwater manipulator based on binocular vision. Based on binocular vision and image recognition, the accuracy of the end of the measuring manipulator is measured. Point cloud matching and Modified-Denavit-Hartenberg (MD-H) parameter identification method based on error model and least square method are used for iteration. The theoretical simulation and experiment based on the four-degree-of-freedom manipulator (4-DOF) show that the proposed detection method can quickly obtain the end position of the manipulator. The iteratively identified M-DH mathematical model improves the positioning accuracy and lays a foundation for the long-term high-precision underwater operation of the underwater manipulator.