The roundness measurement of large circular parts is mainly made by manual contact measurement, which is less efficient. For the disadvantage of contact measurement, the non-contact measurement technology based on machine vision is used to study the roundness measurement of large ring parts without direct contact with the parts to be measured, and the data of roundness and concentricity can be obtained by fitting the partial circular image of parts. At first, preprocessing the acquired image, and aiming at the influence of the noise and other factors, the bilateral denoising algorithm with the best processing effect is selected. And then, based on the traditional Canny edge detection, an improved Canny adaptive edge detection algorithm is developed for edge detection and contour extraction. At last, in the image fitting link, based on the traditional least squares fitting algorithm, the least squares iterative fitting algorithm is used to fit the partial arc of large circular parts, and the validity and superiority of the algorithm are validated. The measuring method is convenient and reliable, which improves the detection efficiency, and can quickly measure the size of the part within the allowable error.