A novel framework to extract contours in ring rolling process

Extracting contours from ring forgings during the dynamic rolling process is fundamental for subsequent tasks such as diameter measurement, roundness evaluation, and geometry reconstruction. Existing methods perform well on static forgings but often fail to extract contours throughout hot deformation, facing a trade-off between accuracy and robustness. To address this, we introduce a novel and low-cost algorithmic framework based on computer vision technologies. It achieves rapid, highly accurate and robust contour extraction. The process begins with semantic segmentation using the lightweight zero-shot model, MobileSAM. The generated forging mask and defined occlusion zones refine edge detection results. Our proposed contour vote algorithm then isolates the visible portions of the inner and outer contours. These segments are completed via a compensation step and integrated into the final contours. The entire operation takes less than 0.1 s, and the contour evaluation score, Optimal Image Scale (OIS), is near 1, suitable for industrial applications. Extensive testing against 324 Canny threshold variations validates the framework's exceptional robustness and precision.