Abstract: Aim at the quality control problems during the process, visual sensing technology was introduced. In order to obtain the visual feature information accurately, the imaging system was analyzed, imaging model was put forward, and the sub-pixel corner extraction algorithm was compared. External parameters and extrinsic parameters was obtained by improved calibration algorithm of Zhang using color optimized calibration target. Results show that calibration error is 0.712 305 pixel, it namely that 0.123 mm. Two dies was repaired after the calibration, which could meet the accuracy requirement and this method was practical and reliable.