Multi-Frame Temporal Integration for 3-D Shape Measurement of Freely Falling Small Objects Using a High-Speed Camera...

Area of Science:

• Computer Vision
• 3-D Reconstruction
• Metrology

Background:

• High-speed 3-D reconstruction is limited by viewpoint availability in fixed camera arrays.
• Insufficient viewpoints lead to incomplete or inaccurate geometry in single-frame multi-view stereo.

Purpose of the Study:

• To develop a multi-frame temporal integration approach for enhanced 3-D reconstruction of moving objects.
• To overcome the limitations of single-frame reconstruction by leveraging the rigid-body assumption.

Main Methods:

• Utilized a three-layer circular array of 32 synchronized RGB cameras (1440x1080, 160 fps).
• Developed a free-fall-oriented algorithm for active frame detection and temporal window selection.
• Integrated accumulated multi-frame images into a structure-from-motion and multi-view stereo (SfM-MVS) pipeline.
• Simultaneously recovered object 6-DOF pose trajectory using SfM-estimated camera parameters.

Main Results:

• Single 32-camera frame achieved an F-score > 0.97 at a 0.5 mm threshold against a structured-light scanner.
• Temporal frame accumulation reached a plateau F-score of 0.984, with diminishing returns beyond 1-2 frames.
• Achieved sub-millimeter accuracy (mean error 0.146±0.033 mm, mean relative error ~1.6%) for 8-10 mm objects.
• Resolved fine surface features like cracks, enabling visual defect identification.

Conclusions:

• Rigid-body multi-frame temporal integration is effective for high-throughput, non-contact 3-D inspection of small, moving objects.
• A small number of temporal frames is sufficient for convergent sub-millimeter accuracy.
• The method enhances viewpoint density without additional hardware, improving reconstruction quality.