Large-scale hole-axis assembly structures are critical components in helicopter lift systems, and their assembly quality directly affects overall performance and service safety.directly impacting overall performance and service safety During visual pose measurement, monocular vision suffers from a limited field of view, complicating acquisition of complete end-face images while maintaining high measurement accuracy, and extending the working distance further degrades local precision. A multi-camera vision-based pose measurement method is proposed. First, a target coplanarity constraint establishes a unified mapping model from multiple camera image planes to the primary camera image plane, achieving geometrically consistent multi-source image representation via plane-induced homography. Second, a discrete arc-segment ellipse fitting method using keypoint search detects arc keypoint pairs through pixel-level jump detection and smooths them with centroid coordinates, incorporating the hyper least squares method to improve the accuracy and robustness of ellipse parameter estimation under incomplete edge conditions. Finally, geometric prior constraints of hole-axis assembly eliminate ambiguity in spatial circle pose estimation, while a weighted optimization method based on forward projection geometric distance iteratively refines the pose parameters. Experiments on a four-camera industrial vision system demonstrate a circle center accuracy of 0.037 mm and a normal vector accuracy of 0.053°, meeting engineering requirements for high-precision assembly of large-scale hole-axis structures. The proposed method provides an effective solution for vision measurement of large-scale components, balancing global coverage with local precision.