An improved Canny operator is proposed through researching and analyzing the data of the grey image with statistical principle on the basis of deep analysis of traditional Canny edge detection operator. Compared with the traditional Canny edge detection operator, improved Canny operator replaces the original grey value of pixels with average value of similar gray level around the edge points and introduces a kind of adaptive method to determine the high and low threshold value. Applying the improved Canny operators to the robot vision system. The experimental results show that the improved Canny operator can balance the relationship between noise and edge information and extract the edges of the measured object better over the traditional Canny operator.