Figure 1
Flowchart of the architecture of PIVlab.
Figure 2
Main graphical user interface. A: Menu with several sub menu entries. B: Main panels are shown in this area, the content changes depending on the menu item that was selected. C: Quick access toolbar, allowing to access the most important menu items. D: Tools panel, showing information about the currently selected point. It also allows to skip through the input images/video, and toggle within an image pair. Zoom and pan buttons are also shown. E: Image area. Images and results are shown here.
Table 1
Correlation robustness settings in PIVlab.
| CORRELATION ROBUSTNESS | WINDOW DEFORMATION | CROSS CORRELATION | REPEATED CORRELATION | PROCESSING TIME |
|---|---|---|---|---|
| ‘Standard’ | linear | circular | off | – |
| ‘High’ | spline | linear | off | ++ |
| ‘Extreme’ | spline | linear | on | ++++ |
Figure 3
The effect of ensemble correlation for synthetic particle images with only 0.64 particles per interrogation area. Left: True displacement of the synthetic particle images. Middle: Average displacement of an analysis of 75 images using regular correlation. Right: Ensemble correlation of 75 images.
Figure 4
Mean absolute bias error for synthetic image pairs with increasing noise and particle pair loss. PIVlab’s ‘extreme’ robustness setting performs best, and PIVlab performs generally better than the commercial software.
Figure 5
Mean RMS error for synthetic image pairs with increasing noise and particle pair loss. For image pairs with low noise and low particle pair loss, the commercial software performs slightly better. At higher noise and particle loss levels, PIVlab’s ‘extreme’ correlation robustness performs best.
Figure 6
The amount of unsuccessful displacement estimates (estimates that deviate more than 1 pixel from the true displacement) serve as an indicator for the robustness of an analysis. PIVlab’s ‘extreme’ correlation robustness performs best, followed by the commercial software.
Figure 7
Computing time per displacement estimate. PIVlab is between 3 and 30 times slower than the commercial software.
Figure 8
Background elimination and auto-correlation suppression in PIVlab. A: PIV data without artificial background (true displacements) B: A background signal is superimposed on the particle image. The signal of the background is stronger than the signal of the moving particles, therefore, a displacement of zero is detected in these areas. C: The background signal has been removed from the images before analysis by subtracting the average intensity of a larger amount of image pairs. The image has become darker in the background area, but a weak particle signal became visible again. Displacement estimates are possible also in the area where a strong background signal was present. D: Auto-correlation suppression ignores displacement estimates that are close to zero and takes the second highest peak in the correlation matrix for estimating the displacement. It can effectively remove a correlation of the background signal in this case. E: Absolute difference between the true displacements in A and the displacements in B. Displacements in the areas of strong background signal are incorrect. F: Absolute displacement difference between A and C. Displacement estimates are correct, removing the background signal does not affect the displacement estimate. G: Absolute difference between A and D. Displacement estimates have only a small difference to the true displacements.
Figure 9
Propeller anemometer in a water tunnel. The flow around the anemometer was mapped using PIV. The mean flow velocity inside the red rectangle was used for the comparison.
Figure 10
Comparison of flow speed measurements with PIVlab and a propeller anemometer. The relation is linear and has a slope of almost unity with almost zero offset.
Figure 11
Textures that were tested in PIVlab. From left to right: Experimental particle image, checkerboard, difference clouds, gaussian noise, combination of checkerboard + difference clouds + gaussian noise.
Figure 12
Bias error of the analysis of different texture types. Particle images perform best with a bias of 0.014%. The ‘difference clouds’ texture has a bias of 5.6%.
Figure 13
RMS error of the analysis of different texture types. Particle images perform best with an RMS error of 0.01%. The ‘difference cloud’ texture has an RMS error of 3.6%.
Figure 14
Unsuccessful correlation estimates in the analysis of different texture types. Gaussian noise images show a low rate of successful displacement estimates.
Figure 15
Correlation matrices for a displacement of 5 pixels. From left to right: Experimental particle image, checkerboard, difference clouds, gaussian noise, combination of checkerboard + difference clouds + gaussian noise.
Figure 16
Principle of the Gaussian fit: Subpixel precision is achieved by fitting a one-dimensional Gaussian function (solid line) to the integer intensity distribution of the correlation matrix (dots) for both axes independently (only one axis is shown here). [Taken from 16].