3D Measurement of Complex Textured Objects Based on Bidirectional Fringe Projection

Yuchong Chen, Jian Yu, Shaoyan Gai, Zeyu Cai, Feipeng DaAAAI 2025 · DOI

Abstract

High-precision 3D measurement of objects with complex textures is often hampered by phase errors caused by lens defocusing at color boundaries. This paper presents the ELPC-PCF method, which utilizes bidirectional fringe projection and epipolar geometry constraints to correct these errors. By merging point clouds from horizontal and vertical projections, the proposed approach reduces measurement errors by over 65% compared to standard techniques, offering a robust solution for industrial inspection of textured surfaces.

Introduction

Fringe Projection Profilometry (FPP) is widely adopted for 3D scanning due to its high speed and resolution. However, its performance degrades significantly when measuring objects with complex surface textures. In real-world setups, lens defocusing is nearly unavoidable, causing pixels near sharp color boundaries to capture a blurred mixture of light. This cross-contamination shifts the recorded phase, resulting in "ghosting" artifacts and jagged edges in the final 3D point cloud. Existing solutions, such as deblurring algorithms or complex calibration models, are often computationally intensive or limited to specific surface geometries.

Methods: The ELPC-PCF Approach

The researchers developed a two-stage method called Epipolar Line Phase Correction and Point Cloud Fusion (ELPC-PCF).

1. Epipolar Line Phase Correction (ELPC)

The core innovation lies in using the physical geometry of the camera-projector system. In a calibrated setup, any point captured by the camera must correspond to a specific epipolar line in the projector's coordinate space. Defocus-induced phase shifts cause reconstructed points to "drift" from these lines. By applying geometric constraints, the ELPC algorithm calculates a correction factor to reposition these points accurately without requiring an explicit estimation of the lens's Point Spread Function (PSF).

Epipolar geometry constraints used for phase correction in the ELPC-PCF method.

2. Bidirectional Projection and Fusion (PCF)

The system projects both horizontal and vertical fringes. This bidirectional approach ensures that if a texture edge is parallel to one fringe direction (minimizing error), the other direction provides a complementary constraint. The Point Cloud Fusion (PCF) step merges the data from both directions, effectively averaging out systematic errors and suppressing random noise.

Experimental Results

The ELPC-PCF method was evaluated using a custom FPP system on various objects, including calibration plates, curved vases, and textured consumer products. Quantitative analysis on a standard calibration plate showed a reduction in Mean Absolute Error (MAE) by 65.74% and Root Mean Square Error (RMSE) by 67.75% compared to uncorrected results.

Comparison of 3D reconstruction error maps for a textured mouse: (a) original image, (b) uncorrected error map, and (c) corrected error map using ELPC-PCF.

As shown in the figure above, the uncorrected error map (b) exhibits high-intensity noise and inaccuracies at the boundaries of the object's features. The ELPC-PCF method (c) significantly smooths these regions, producing a much more accurate 3D representation. The method also outperformed the state-of-the-art Two-Dimensional Error Model (TEM) by approximately 30% in accuracy across complex curved geometries.

Conclusion

The ELPC-PCF method provides a significant leap in 3D measurement accuracy for textured objects by leveraging geometric constraints rather than complex image processing. While the requirement for bidirectional projection doubles the acquisition time, the substantial gain in precision and robustness makes it highly suitable for industrial quality control. Future work will focus on reducing the frame count and exploring deep learning integration to further optimize the correction process.

Summary updated

BibTeX
@inproceedings{chen2025d,
  author = {Yuchong Chen and Jian Yu and Shaoyan Gai and Zeyu Cai and Feipeng Da},
  title = {{3D Measurement of Complex Textured Objects Based on Bidirectional Fringe Projection}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2025},
  doi = {10.1609/aaai.v39i2.32233},
}
Figure