The kinematic parameter identification technique of robot arm is effective for enhancing the absolute positioning accuracy. For the low efficiency of the general calibration methods based on planar constraints, which data can only be captured by manual teaching method, an automatic method was proposed for the kinematic parameter identification of the robot arm with the help of a stereo vision system to locate the constraint planes. In order to remedy the limitation of the common field range of two industrial cameras, three target spots were attached to each constraint plane, so the location of each constraint plane could be equivalent to locate their target spots correspondingly. The binocular stereo vision system extracted the center of target spots and then matched them respectively, so that the 3D position information of each target spot in the base coordinate system of the robot arm was calculated. Meanwhile, a target coordinate system was built to plan the constraint points according to a certain rule. To enhance the calibration precision further, an error model based on two planes perpendicular to each other was proposed, then a set of nominal vectors of each plane were calculated by three non-collinear constraint points and the dot product of each pair of nominal vector was zero, namely, there existed an extra restraint in the error model. Then a probe fixed to the end flange of robot arm was used to touch two perpendicular constraint planes automatically, and the actual values of all kinematic parameters could be identified by the improved least squares. Experiments showed that the absolute positioning accuracy of the robot arm was enhanced from 1.234 mm to 0.453 mm after modifying the kinematic parameters based on the proposed error model. The proposed method realizes automatic measurement of data, and highly increases the efficiency of robot arm calibration, and provides a reference for mass calibration, which has great significance of engineering.