Abstract
In practical particle image velocimetry (PIV) applications, severe noise and dynamic optical artifacts—such as sharp shadows, intense specular reflections, transient occlusions, and non-uniform illumination—critically degrade particle-to-background separation and yield regions with highly unreliable correlation peaks. Under these extreme conditions, dynamic mask generation is essential to restrict the interrogation to the physically valid domain and to suppress spurious vectors. This work introduces a robust dynamic masking method based on the local particles-per-pixel (ppp) distribution, derived entirely from a single frame. This single-frame architecture makes the method applicable even when stationary-background assumptions are untenable or long time series are unavailable. The proposed pipeline comprises (i) an optional algorithmically simple pre-processing, (ii) tracer identification, (iii) spatial aggregation to construct the ppp field, (iv) smoothing and segmentation, and (v) morphological corrections for boundary refinement. A controlled synthetic analysis isolates typical degradation mechanisms and provides an objective basis for parameter selection. Finally, the approach is stress-tested on three highly noisy, dynamic experimental cases: a pediatric ventricular assist device, a multiphase bubble column, and the undulating boundary of an anguilliform swimming fish. The proposed pipeline is quantitatively benchmarked against the standard peak correlation energy (PCE) masking approach. Results demonstrate that while the PCE approach fails in regions of severe shadowing and continuous boundary deformation, the single-frame ppp method systematically isolates these optical failures. Evaluations of the probability density functions of the primary peak ratio confirm that the proposed mask substantially reduces low- signal-to-noise ratio spurious vectors, significantly enhancing global correlation fidelity while maintaining stable frame-by-frame application after parameter selection.

This publication has 23 references indexed in Scilit: