A Machine Vision-based Dimension Measurement Method of Shaft Parts with Sub-pixel Edge

Abstract The evolution of the automotive industry demands gearboxes that offer improved specifications in terms of power transfer and shifting performance. The high-quality standards required for transmission component manufacturing processes render it necessary to measure shaft parts. To realize online diameter inspection of shaft parts in the gearbox production line, a machine vision-based non-contact metrology method with Canny-Steger subpixel edge detection is proposed. The interference background was separated from the measured target using ROI technology. Morphological closure operations and bilateral filtering preprocessing were performed on the collected images to achieve filtering and denoising, respectively. With the extraction of pixel-level edges using the Canny operator as a basis, the Steger algorithm was used to detect the subpixel edges. The edges and normal directions are derived from the Hessian matrix. Subsequently, Taylor polynomials are applied to the pixel grayscale in the normal direction to obtain the grayscale distribution function, the extreme values of which are solved to obtain the subpixel edge point positions. The fitting data are employed to evaluate the machining errors of the parts, significantly reducing redundant calculations in the Steger algorithm process and improving the detection speed. The proposed method was validated through synthetic and real experiments. Experimental results demonstrate that the measurement method proposed in this paper can rapidly measure shaft parts. Its diameter measurement accuracy can reach 3μm, and the repetitive measurement accuracy can reach 2μm. Compared with pixel-level edge detection, the accuracy has been improved by 60%.