Based on the computer vision system, the methods of measuring the teeth number and the length of common normal line of gear was put forward for the first time in gear detection. The polar coordinate transformation algorithm was first used to transform the preprocessed tooth profile sampling data. The tooth profile curve on the circumference was transformed into a horizontal state, and the obtained tooth profile regarded as a sinusoidal curve. The Fourier transform function in the Matlab toolbox was employed to get the fitting expression of the sine curve, and then the ratio of the total number of tooth profile sampling data (i.e., the number of columns) and the period of the fitting function was rounded to obtain the number of teeth of the detected gear. As for the dection of changes of common normal line length, first, the radius and modulus of the tooth tip circle were obtained to get the radius of the base circle, thereby obtaining midpoint and slope at the intersection of the base circle and the tooth profile. The tangent equation tangent to the base circle was obtained by using the point slope equation, and the length of the intersection of the tangent line and the tooth profile spanning k teeth was whose length of the common normal, the length variation whose was obtained according to the difference between the maximum value and the minimum value. By the methods proposed above, the non-contact accuracy detection of gear can be realized, whose accuracy can meet the needs of engineering practice.