Vibration Detection and Motion Compensation for Multi-Frequency Phase-Shifting-Based 3D Sensors

Liya Han, Zhongwei Li, Kai Zhong, Xu Cheng, Hua Luo, Gang Liu, Junyun Shang, Congjun Wang, Yusheng ShiSensors 2019 · DOI

TL;DR

This paper introduces a novel, image-based method for detecting and compensating for vibrations in multi-frequency phase-shifting 3D sensors. By analyzing image sequences, the method quantifies vibration intensity and corrects motion-induced errors, enabling more accurate 3D reconstructions in challenging industrial environments.

What Problem Does This Paper Address?

Industrial in-situ 3D inspection, often performed on production lines, relies on optical methods like phase-shifting profilometry (PSP). However, these systems are highly susceptible to environmental vibrations, which are common in manufacturing settings like forging factories. These vibrations can introduce significant errors into the 3D data, leading to faulty inspection results. While physical vibration sensors exist, they often struggle to detect the low-frequency vibrations prevalent in these environments. Existing image-based vibration detection methods, often relying on image blur, are not directly applicable to the artificial edges created by fringe projection. Therefore, there's a critical need for an effective, image-based method to detect and compensate for vibrations specifically within multi-frequency PSP systems.

Methods

The core of the proposed method involves two main components: vibration detection and motion compensation.

Vibration Detection: The authors propose an image-based metric to quantify vibration intensity. This metric leverages the fact that in multi-frequency PSP, the system relies on precise pixel correspondence between frames captured at different frequencies. Vibrations disrupt this correspondence. The method analyzes the "streak component" in a "virtual frame" (a uniformly illuminated image derived from the superposition of fringe images). The intensity of this streak component, detected via Fourier transform analysis, is directly correlated with the magnitude of the vibration. For situations where the streak component is too weak to detect, an alternative metric, the L1-norm of feature point differences between virtual frames, is used.

Motion Compensation: Once vibration is detected and its intensity quantified, the method compensates for the resulting motion error. This is achieved by correcting the pixel correspondence between different wrapped phase maps. The key idea is to use a homography matrix, calculated from feature points detected in virtual frames, to map one wrapped phase map to another. This process effectively "repairs" the damaged pixel correspondence caused by vibration, allowing for accurate phase unwrapping and subsequent 3D reconstruction. The method retains the pixel-wise isolation principle of PSP, which is crucial for handling complex object surfaces.

Datasets

Not discussed in the paper. The experiments were conducted using a self-built 3D sensor and various objects (paper, industrial parts) to simulate vibration conditions.

Key Insights and Ideas

  • Vibration's Impact on Multi-Frequency PSP: The paper clearly illustrates that vibrations primarily disrupt the pixel correspondence between different frequencies in multi-frequency PSP, rather than solely affecting individual phase-shifting subsequences. This leads to "broken" surfaces in the 3D reconstruction.
  • Image-Based Vibration Metric: The use of the "streak component" in a virtual frame as a vibration indicator is a novel approach. It directly links a measurable image artifact to the underlying physical vibration.
  • Homography for Correspondence Correction: Applying homography, typically used for planar transformations, to correct pixel correspondence between wrapped phase maps is a clever adaptation. It leverages the fact that for low-parallax situations (typical in many industrial inspections), homography can accurately map image transformations.
  • Prioritizing Vibration Detection: The paper emphasizes that accurate homography calculation requires sufficient feature points and a low enough vibration intensity. Their vibration detection metric acts as a crucial prerequisite for successful compensation.
  • Dual Metrics for Robustness: Employing both streak intensity and L1-norm/dpixel metrics provides a more robust approach to vibration detection, covering different intensity ranges and potential failure modes.

Key Results

  • The proposed method successfully detected vibrations and compensated for motion errors in simulated industrial scenarios.
  • The "broken" surfaces in 3D reconstructions caused by vibration were significantly "repaired" after compensation.
  • A comparison of compensated 3D data with static 3D data showed average upper and lower deviations of +0.055 mm and -0.054 mm, respectively, indicating high accuracy.
  • The study identified thresholds for successful compensation: an NL1 value below 3.210, a decreased NL1 below 21.3%, and a dpixel below 0.121.
  • Streak intensity was found to be a looser but still effective indicator of vibration intensity compared to NL1, especially at higher motion speeds.

Limitations and Future Work

  • Feature Point Dependency: The compensation method's success relies heavily on the quality of feature point extraction, matching, and filtering. This limits the algorithm's robustness.
  • Homography Accuracy: The effectiveness of the homography matrix is dependent on the scene's characteristics and the motion intensity.
  • Future Work: The authors plan to improve feature point detection, matching, and filtering algorithms. They also aim to explore alternative methods for acquiring homography matrices beyond grayscale feature points and to optimize the homography matrix using phase information.

Summary updated

BibTeX
@article{han2019vibration,
  author = {Liya Han and Zhongwei Li and Kai Zhong and Xu Cheng and Hua Luo and Gang Liu and Junyun Shang and Congjun Wang and Yusheng Shi},
  title = {{Vibration Detection and Motion Compensation for Multi-Frequency Phase-Shifting-Based 3D Sensors}},
  journal = {Sensors},
  year = {2019},
  doi = {10.3390/s19061368},
}
Figure