Three-Dimensional Measurement Method of Four-View Stereo Vision Based on Gaussian Process Regression

Area of Science:

• Computer Vision
• Metrology
• Machine Learning

Background:

• Single-sensor systems have limited measurement ranges.
• Multisensor systems offer broader ranges but face calibration and data fusion challenges.
• Accurate 3D reconstruction is crucial for various scientific and industrial applications.

Purpose of the Study:

• To propose a novel three-dimensional (3D) measurement method using four-view stereo vision and Gaussian process (GP) regression.
• To develop an optimized data fusion technique for point cloud data.
• To demonstrate superior measurement accuracy and reconstruction effects compared to traditional methods.

Main Methods:

• Acquired two sets of point cloud data using gray-code phase-shifting.
• Designed composite kernel functions for initial GP model creation.
• Implemented a Bayesian inference-based data fusion method with weighted optimization for noise differences.

Main Results:

• Simulations on curves and high-order surfaces showed improved accuracy.
• Experiments on complex 3D objects demonstrated superior reconstruction effects.
• The proposed GP regression method outperformed traditional techniques in measurement accuracy.

Conclusions:

• The proposed four-view stereo vision method with GP regression offers enhanced 3D measurement capabilities.
• The Bayesian inference-based data fusion effectively handles noise variations in point cloud data.
• This approach provides a robust and accurate solution for 3D reconstruction without strict hardware constraints.