Abstract
Precise and real-time acquisition of aircraft model attitude is fundamental for aerodynamic analysis in wind tunnel experiments, yet achieving high-precision non-contact measurement remains a significant challenge. To address this, this paper proposes a pose measurement framework based on a monocular Time-of-Flight (ToF) camera that fuses keyframe global registration with non-keyframe local registration. First, a novel hand-crafted local feature based on three-plane encoded height and density is introduced. When combined with the Two-stage Consensus Filtering RANSAC (TCF-RANSAC) algorithm, this feature achieves robust global registration of keyframes, providing reliable initial pose estimates for the system. Subsequently, leveraging the continuity constraint of model motion, fast incremental local registration of non-keyframes is performed using the Generalized Iterative Closest Point (GICP) algorithm, which avoids falling into local optima while significantly improving computational efficiency. Evaluation results on simulated datasets with synthetic noise and a real experimental platform demonstrate that the method achieves a single-axis rotation angle error of less than 0.03∘ while processing at over 40 FPS, satisfying real-time measurement requirements. Comparative evaluations against multiple existing registration methods indicate that the proposed framework achieves superior accuracy and robustness, reducing rotation angle errors by 9% to 39% compared to mainstream global registration methods under single-view ToF sensing conditions. Furthermore, this study quantifies the error distribution characteristics of monocular ToF-based pose estimation, revealing an "axis-sensitivity" phenomenon where the rotation error around the optical axis is significantly lower (e.g., 0.02∘, 0.03∘) than that around the orthogonal axes (e.g., 0.38∘, 0.26∘). These findings provide practical guidance for camera placement and system design in high-precision aerodynamic measurement scenarios.