In the process of workpiece circular arc radius measurement on binocular vision, the circular arc contour feature extraction is the key to match follow-up edge contour point and reconstruct space arc. Affected by the surface texture of the workpiece, the interference of the surrounding environment, and uneven illumination, the existing algorithm cannot accurately extract the circular arc contour feature. Based on the edge detection by Sobel operator, we use adaptive convolution operation and double local binary pattern texture feature to generate the fusion gray value, which can be used to screen out the edge contour points in the previous detection result by Sobel operator. The polar coordinate distribution histogram is obtained and processed as a normal distribution, which can eliminate the interference of noise points in the complex background and distinguish the outer contour feature from the inner contour feature. The experimental results show that the proposed algorithm overcomes the influence of illumination and becomes accurate and robust.