Abstract: Calibration is a premise of visual computing for robot.According to the particularity of mobile welding robot's workplace, a method for the auto-calibration of hand-eye binocular vision system based on OpenCV is given.Corners are extracted automatically, effective images are selected by itself, internal parameters are calculated based on the extracted result of corners with sub-pixel precision, geometry matrix between CCDs in pair is calibrated via images in pair captured at the same time, the motion matrix of sensor is got indirectly by computing the changed motion matrix of calibration board in the coordinate of sensor, then hand-eye relationship is caculated by least-square method.Calibration results demonstrate that the method can meet the calibration precision for vision system of mobile welding robot used for intelligent welding, meanwhile improve the autonomy because there is no human intervention compared with the traditional Matlab calibration toolbox.