A Precision Measuring Approach of Shaft Edge using Dual-stage Sub-pixel Detection with Lighting Compensation

This research introduces a method with a pragmatic approach for measuring the contour of shaft workpieces. The algorithm utilizes dual-stage Canny edge detection and sub-pixel localization techniques. This study focuses on the challenges associated with edge detection in the context of back-light lighting systems, with the aim of enhancing both accuracy and stability. The proposed technique integrates adaptive threshold-based Canny edge detection, weighted fusion, and morphological algorithms in order to improve the continuity of edges. The process of identifying the location of a coarse edge is accomplished through the utilization of cubic spline interpolation and curve fitting techniques to determine sub-pixel coordinates. The RANSAC fitting algorithm is employed to estimate lines that best match a set of edge points, hence enabling the computation of accurate values for axis-type components. The error analysis and compensation under various lighting conditions were completed through experimental analysis. The method described in this study demonstrates a notable level of accuracy and reliability in the recognition of cylindrical standard parts. It exhibits the capability of precisely identifying standard parts of varying sizes, with an absolute error margin of 8 micrometers. In addition, the algorithm has anti-interference capability, and we deployed the algorithm on Nvidia's Jetson orin nano, so that the entire system design can be more compact. In conclusion, this algorithm offers a viable solution for the efficient and accurate recognition of part contours in diverse settings.