Towards in-situ monitoring of floating platforms: geometry-anchored recurrent diffusion for long-sequence motion reconstruction based on monocular marker-free video

Document Type

Article

Publication Date

9-1-2026

Abstract

Long-sequence motion reconstruction of floating platforms is important for response-informed monitoring because downstream tension estimation and fatigue-related assessment depend on temporally consistent motion histories rather than isolated pose estimates. In practical offshore applications, direct and accurate long-term in-situ measurement is especially difficult, since the monocular marker-free vision is constrained by limited geometric observability, full-sequence label unavailability, and growing drift risk over extended durations. To address these challenges, this study develops a new response-informed vision-based monitoring framework for floating platforms under monocular, marker-free observation. The framework combines geometry-anchored pseudo-supervision, recurrent diffusion-based temporal inference, stabilization-oriented long-sequence tracking, and downstream tension and damage-equivalent-load (DEL) calculation under an adopted simplified mooring model. The method is evaluated through an OpenFAST-based primary case, ablation studies, a supplementary-view analysis, a controlled wave-tank test demonstration, robustness evaluations under changed excitation and controlled visual disturbances, and a runtime assessment. The results show that the dominant platform motion can be reconstructed with stable long-sequence behavior, while recurrent temporal inference and reliability-aware stabilization play complementary roles in preserving sequence continuity and suppressing drift accumulation. By integrating learning-based efficient reconstruction with stabilization-oriented temporal inference, the proposed framework addresses a critical methodological gap in achieving computationally efficient and temporally consistent motion reconstruction over long sequences, even under visual degradation. This capability demonstrates the potential for in-situ offshore infrastructure monitoring. Under the adopted quasi-static mooring model with axial elasticity, the reconstructed motion preserves the main information required for the mooring tension and downstream DEL estimation, facilitating the fatigue-related assessment.

Publication Title

Mechanical Systems and Signal Processing

Link to Full Text

Share

COinS