Neural-network method applied to the stereo image correspondence problem in three-component particle image...

Area of Science:

• Fluid dynamics
• Computational intelligence
• Image processing

Background:

• Stereoscopic particle image velocimetry (PIV) is crucial for fluid flow analysis.
• Stereo image-pair reconciliation is a key challenge in PIV, especially in tracking mode.
• Traditional methods face difficulties with high particle densities and complex flow patterns.

Purpose of the Study:

• To introduce and evaluate a recurrent neural network (Hopfield type) for stereo image-pair reconciliation in PIV.
• To demonstrate the effectiveness of this neural network approach on both synthetic and real-world flow data.
• To establish a quantitative performance metric (PPID) for assessing the reconciliation method.

Main Methods:

• Application of a Hopfield-type recurrent neural network for matching particle image pairs across stereo views.
• Definition and utilization of the partner-particle image-pair density (PPID) parameter.
• Testing the network on virtual-flow and physical-flow PIV datasets, including hydraulic flow.

Main Results:

• The neural network successfully reconciles stereo image pairs in PIV.
• Pairing accuracy remains above 97% for PPID values up to 5, starting at 100% for PPID of 1.
• Demonstrated successful application to a hydraulic flow dataset, validating its real-world utility.

Conclusions:

• The Hopfield-type recurrent neural network offers a robust and accurate solution for stereo PIV image analysis.
• The PPID parameter provides a valuable metric for quantifying the challenge and performance of PIV reconciliation.
• This novel neural network approach significantly advances the capabilities of stereoscopic PIV.