Automatic reading recognition of index instruments based on computer vision is studied in this paper according to the precision demand of the calibration system. The panel center can be obtained by the total least squares method. Then the skeleton of the pointer can be obtained with the panel center known. At last, an improved angle method based on the angle method and the distance method is proposed to recognize the gauge reading. The process of obtaining the panel center and the skeleton of the pointer produces a precise outcome and a small amount of calculation, while the improved angle method combines the advantages of the angle method and those of the distance method to increase the precision further.