A novel sub-pixel measurement method for industrial image processing applications

Industrial image processing systems are widely utilized in high-precision fields such as the automotive, medical, and aerospace industries. Although specialized equipment such as telecentric lenses is used for precise measurements, mechanical and software-related errors may still occur when measuring parts of different diameters within the same system. This dissertation proposes a subpixel counting-based method to enhance measurement accuracy, along with conversion factor-based and pixel-based diameter estimation methods. The subpixel counting method has been developed to improve the accuracy of diameter measurements for ring-shaped objects. In this method, image pixels are classified into full pixels (completely contained within the object) and transition pixels (gray-scale pixels located at the boundary between the object and its background). To enhance the contribution of transition pixels to diameter calculations, a normalization process is applied. The diameter is computed using the circle area equation, while a refined thresholding method ensures the accurate classification of gray-scale pixels. Furthermore, the conversion factor-based and pixel-based estimation methods proposed in this study enable the high-precision measurement of objects with varying diameters. In the conversion factor-based approach, the diameter is estimated by determining the conversion factor from known reference parts, while in the pixel-based approach, the diameter is directly estimated based on the pixel measurements of reference parts, eliminating the need for additional conversion steps. The motivation for this study arises from the need for accurate measurement of parts with different dimensions using the same mechanism while avoiding the high costs associated with producing dedicated reference gauges. To address these challenges, this research introduces a generalizable approach that operates independently of specific measurement algorithms. Experiments conducted using industrial cameras and telecentric lenses demonstrate that the proposed methods significantly reduce measurement errors. Initially ranging between 13–114 µm, measurement errors were reduced to 1–2 µm with the proposed approaches. The operational sensitivity of the subpixel counting method was determined to be 1/20th of the pixel size, with an average uncertainty of 1 µm. Additionally, the proposed algorithm achieved 3–10% higher accuracy compared to existing methods and improved computation time by 12.5–35%. This study introduces an innovative algorithm to achieve subpixel-level accuracy in diameter measurement, making a significant contribution to the existing literature. The proposed methods enable high-precision measurement of all parts within the camera's field of view using only a limited number of known reference parts, thereby reducing error rates and improving measurement reliability.