A ROS-Based Multi-Camera Framework for Real-Time 3-D Structural Displacement Tracking

초록

Structural Health Monitoring (SHM) requires precise displacement measurements to ensure infrastructure safety and longevity. Vision-based methods offer a scalable, non-contact solution but are often constrained by synchronization and real-time processing challenges. This paper presents a Robot Operating System (ROS)-based framework for real-time 3D displacement tracking that leverages software-synchronized multi-camera systems and AprilTag fiducial pose estimation. The framework makes three tightly coupled contributions: 1) an end-to-end software synchronization pipeline that integrates host clock discipline, SDK capture-time timestamp propagation, and middleware tuple formation for real-time cross-host camera alignment without specialized hardware or post-hoc alignment; 2) a scalable architecture for distributed, multi-point tracking; and 3) a modular real-time pipeline enabling parallel acquisition and extensibility. In the primary LDS-referenced displacement validation experiments, the system achieved sub-millimeter accuracy for the validated in-plane displacement components (RMSE 0.14–0.40 mm) and maintained residual timing offsets below 5 ms across the tested configurations. Additional repeated-run LDS-referenced sinusoidal tests further confirmed repeatable best-view performance across excitation frequencies, with best-view mean RMSEs ranging from 0.096 to 0.261 mm. Subsequent modal testing verified relative frequency errors under 4% across all modes and MAC values exceeding 0.85 for the first four modes. These results demonstrate that consumer-grade cameras, when combined with software-only synchronization, can provide laboratory-grade precision in compact, field-oriented setups for real-time displacement tracking.

키워드