A Vision-Based Procedure with Subpixel Resolution for Motion Estimation

Area of Science:

• Structural Engineering
• Computer Vision
• Signal Processing

Background:

• Vision-based motion estimation for structural systems is a growing field.
• Accurate motion estimation for structures, especially with varying amplitudes, remains challenging.
• Existing methods often struggle with computational efficiency and accuracy for diverse motion scales.

Purpose of the Study:

• To propose a robust, multi-step framework for vision-based motion estimation in structural systems.
• To accurately estimate both large and small motion amplitudes with high precision.
• To extract structural modal parameters (mode shapes, vibration frequencies) from video data.

Main Methods:

• A hybrid approach combining stochastic (population-based block matching) and deterministic (Reduced-Error Gradient) methods.
• Smart Block Matching with Reduced-Error Gradient (SBM-REG) for integer and subpixel motion estimation.
• Complexity Pursuit, a blind source separation technique, for output-only modal analysis.

Main Results:

• The SBM-REG method demonstrated effective motion estimation across various amplitudes.
• Structural mode shapes and vibration frequencies were successfully extracted from video recordings.
• Validation against synthetic data, a cantilever beam, and laboratory tests confirmed the method's accuracy and efficiency.

Conclusions:

• The proposed SBM-REG framework offers a powerful and accurate solution for vision-based structural motion estimation.
• This approach enables reliable extraction of modal properties from video, advancing structural health monitoring.
• The hybrid strategy effectively balances computational load and precision for real-world structural analysis.