<p>High-precision ring forgings produced via the ring rolling process are critical components in demanding sectors such as aerospace. In advanced intelligent manufacturing, automated extraction of contours during this process is paramount for ensuring the geometric accuracy of high-performance rings. However, current industrial practice predominantly relies on inefficient and hazardous manual inspection. Direct application of existing computer vision techniques is impeded by several challenges: the irregular deformations inherent to the process, partial occlusions by mechanical components, and the prohibitive cost of acquiring large-scale annotated datasets for supervised models. To address these challenges, we propose a two-stage machine vision framework for the accurate and rapid extraction of the inner and outer contours on a deforming ring’s upper surface. The first stage integrates a lightweight zero-shot segmentation model with classical computer vision algorithms to precisely label the initial target contours on a static ring. The second stage performs continuous tracking, centered around our proposed optical flow-guided Pointwise Predictor. By expanding the search area for each contour point within defined hyper-parameters, a mechanism that uniquely overcomes the critical issue of cumulative error accumulation inherent in simpler optical flow-based methods, thereby effectively preventing tracking failure without sacrificing precision. Through comprehensive experiments, including ablation studies, detailed hyper-parameter analysis, and cross-dataset validation on rings of four different colors, we demonstrate the framework’s high performance (ODS F-measure &gt; 0.98), strong generalizability, and exceptional temporal stability. Our work presents a practical and effective solution for automating in-situ process monitoring in the ring rolling industry.</p>