ing at the interference of the surface burrs, oil, and other attachments on the corner detection in the visual measurement, we propose a method of sub-pixel location based on curvature and gray. Firstly, we use a morphological and bilateral filtering method to eliminate burrs, oil, and other attachments in the region of interest. Secondly, we detect the candidate corners according to the curvature characteristics, pre-screen the false-corners by the multi-scale invariance of the curvature angle at the corner, and use the gray information in the circular window to further eliminate false-corners to achieve the corner of the rough positioning. Finally, we screen the edge point of the original image according to the connection between the coarse positioning corner and the regional end points, and fit the filtered edge points by using least squares fitting to achieve precise positioning of the corner point. The experimental results show that the method can effectively overcome the interference of the surface attachment of the workpiece surface. The repeatability of corner location reaches 0.01 mm, and the accuracy of corner location algorithm reaches 0.004 mm, and the comprehensive measurement accuracy based on corner points is 0.06 mm.