Research on a Camera Position Measurement Method based on a Parallel Perspective Error Transfer Model

Camera pose estimation from sparse correspondences is a fundamental problem in geometric computer vision and remains particularly challenging in near-field scenarios, where strong perspective effects and heterogeneous measurement noise can significantly degrade the stability of analytic PnP solutions. In this paper, we present a geometric error propagation framework for camera pose estimation based on a parallel perspective approximation. By explicitly modeling how image measurement errors propagate through perspective geometry, we derive an error transfer model that characterizes the relationship between feature point distribution, camera depth, and pose estimation uncertainty. Building on this analysis, we develop a pose estimation method that leverages parallel perspective initialization and error-aware weighting within a Gauss-Newton optimization scheme, leading to improved robustness in proximity operations. Extensive experiments on both synthetic data and real-world images, covering diverse conditions such as strong illumination, surgical lighting, and underwater low-light environments, demonstrate that the proposed approach achieves accuracy and robustness comparable to state-of-the-art analytic and iterative PnP methods, while maintaining high computational efficiency. These results highlight the importance of explicit geometric error modeling for reliable camera pose estimation in challenging near-field settings.

💡 Research Summary

This paper addresses the long‑standing difficulty of estimating camera pose from sparse 3‑D‑2‑D correspondences in near‑field scenarios, where strong perspective distortion and heterogeneous measurement noise can severely degrade the stability of both analytic and iterative PnP solutions. The authors introduce a geometric error propagation framework that explicitly models how image‑space errors are transferred through the perspective projection as a function of feature layout, camera‑to‑target distance, and the depth‑dependent noise characteristics of each observation. By adopting a parallel‑perspective approximation—essentially a first‑order linearization of the full perspective model valid when the target occupies a large portion of the field of view—the framework yields a closed‑form error‑transfer model (the “error transfer model”). This model quantifies three key factors: (i) the spatial distribution of feature points (central versus peripheral), (ii) the absolute depth of the target (shallower depths amplify image noise more strongly), and (iii) the non‑uniformity of measurement noise across the image (e.g., caused by illumination gradients or underwater scattering).

Building on this analysis, the authors propose a two‑stage pose estimation pipeline. In the first stage, a parallel‑perspective solution provides an initial estimate of rotation and translation that is intrinsically closer to the true solution than the traditional weak‑perspective initialization used in many EPnP‑based methods. In the second stage, the error‑transfer model is used to compute per‑point weights that reflect each observation’s expected contribution to pose uncertainty. These weights are incorporated into a Gauss‑Newton optimization loop, effectively down‑weighting noisy peripheral points while emphasizing well‑conditioned central points. The resulting algorithm retains the O(n) computational complexity of EPnP, yet demonstrates markedly improved convergence behavior and robustness, especially when the number of correspondences is small or when the target is within 0.5 m of the camera.

Extensive validation is performed on synthetic datasets with controlled depth, noise, and feature layouts, as well as on real‑world image sequences captured under three challenging conditions: strong solar illumination (space‑related), surgical lighting, and low‑light underwater environments. Across all tests, the proposed method achieves pose errors comparable to or better than state‑of‑the‑art analytic (EPnP, OPnP) and iterative (POSIT, OI) approaches, often reducing position error by 15‑30 % and rotation error by a similar margin, while maintaining real‑time performance (≈30 fps on a standard CPU).

The paper’s contributions are threefold: (1) a rigorous geometric error‑propagation framework that reveals depth‑ and layout‑dependent instability in near‑field PnP problems; (2) a systematic study of how feature point configuration influences pose stability, providing practical guidelines for marker design in close‑range applications; and (3) a novel parallel‑perspective‑based pose estimator with error‑aware weighting that delivers both high accuracy and computational efficiency. The work underscores the importance of explicitly modeling measurement error in the geometry of near‑field vision systems and opens avenues for more reliable pose estimation in demanding domains such as space docking, underwater vehicle docking, and autonomous surgical robotics.