Version 23 (modified by vaillant1p, 10 years ago) (diff)


Tutorial / Advanced Particle Imaging Velocimetry

This is an example using more advanced tools than the simple example Tutorial: Particle Image Velocimetry?. Image pre_processing and merging of mutiple fields is used to optimise PIV in the case of a narrow parietal jet (produced by convection in a cavity). Open the example UVMAT_DEMO6_PIVconvection (accessible on


Perform the geometric calibration by marking the four box corners with the mouse, as described in Tutorial: geometric calibration?, section 'calibration with reference points'. Alternatively, you can skip this operation by using the reference file Dalsa1.ref.xml provided, removing the extension '.ref' to make it active.


Open the original image with uvmat, selecting the transform option 'phys'. Create a mask polygon by the menu bar command Projection object/mask_polygon. Set the option 'mask_outside' and introduce the coordinates of the four corners in Coord (like for geometry calib). Plot the polygon then create the corresponding mask by Tools/make mask. The default name is 'mask_1.png' in the subfolder Dalsa1.mask.


We observe parasitic light rays on the images which correspond to fixed features, leading possibly to spurious velocity vectors equal to 0. To eliminate those we use sub_bacground. Open the image with 'series' and select the program 'sub_background'. This function is not provided in the default menu, so you need to use the last menu option 'more...', and select the function in the sub-folder 'series/' of the package uvmat. This option is then preserved in the menu for later use. Then run sub_background over the whole index range in i and j, using the default parameters. Answer Yes to the question 'apply levels', which will conveniently rescale the image brightness after background removal.

First PIV

Do PIV on the the whole image series, selecting all the options from Civ1 to Patch2 (see Tutorial: Particle Image Velocimetry? for an introduction).

Choose the pair 'j=1-2' which provides the smaller time interval (100 ms), a good choice to capture correlations in a first try (although higher precision can be obtained with a larger time interval if the correlation is still of good quality).

Select the Mask 'mask_1.png' previously created.

Looking at the results with uvmat, we global flow fetures. We notice hower that the thin parietal plume is not properly resolved while the precision in the interior is poor because of the small displacement within the frame pair. So the next step will be to perform two PIV series optimized for each sub-region and merge them.

Making two masks

The parietal jet requires a very good resolution, particularly among x. To limitate PIV to this jet, let us create a specific mask. Open the previously created contour polygon contour_mask.xml in uvmat by the menu bar Tools/Projection object/browse.... Check the box edit (tag [CheckEditObject]) in the frame Object of uvmat (right hand side) to allow editing of the polygon then replace the lower x bound 0 by 52. Create the corresponding mask by Tools/make mask, save it with name Dalsa1.mask/mask_plume.png. Similarly create a mask for the bulk Dalsa1.mask/mask_bulk.png, with bounds in x [0 55].

PIV on the parietal plume

Choose the following parameters:

  • pair j=1-2 for Civ1 and Civ2 : it minimises the time interval which is needed to capture the large velocity in the plume.
  • CorrBox x,y=(5 31) which optimizes the resolution in x (5 pixels)
  • Search x, y=(25 55) which allows for a larger velocity component along y .
  • auto-grid Dx=2, Dy=10, consistent with CoorBox (about half the box size).
  • Mask selected, with name .../Dalsa1.mask/mask_plume.png .
  • For Civ2, same parameters CorrBox, auto-grid and Mask.
  • For patch and Fix, default parameters.

PIV on the interior

Select the pair j=1-3 to deal with the small velocities (considering that the parietal plume has been masked). Then use the default parameters.

Merging data on a common grid

create a projection grid in phys coordinates. For that open a velocity field with uvmat, displayed in phys coordinates. Use the upper bar menu option Projection object/plane. Then in the GUI set_object, choose the option ProjMode=interp_lin. Choose a mesh 0.1 cm in each direction, ranging from 0 to 58.8 in x and 0 to 55 in y. Press REFRESH to see the result of projection in the GUI view_field. Check the option nb_vec/4 to reduce the number of vectors displayed on the plot.

Now in series open the PIV file Dalsa1.sback.civ_bulk as input. Then append the second file Dalsa1.sback.civ_plume using the menu bar selection Open/Browse? append....

Set FieldTransform to 'phys', and select Projection Object. The plane for projection is then incorporated in series.

It is also advised to introduce masks in the interpolation process so that each field is interpolated in its range of validity. This is done by selecting the option Mask. Use the browser to fill the table of masks, in accordance with the table of input file series.

Merging data using thin plate shell

The previously used linear interpolation does not provide field derivatives. For that purpose, we proceed as previouisly but use the option ProjMode ='interp_tps' for the projection plane in set_object.

In the GUI series, select simultaneously the fields vec(U,V), curl(U,V) and div(U,V) to get the vorticity and divergence in addition to velocity. The calculation is significantly longer that for interp_lin, so in the demo we use a resolution DX=DY=0.2 cm for the projection plane (instead of 0.1 cm).

Open the resulting files with uvmat. Select a vector (components U, V) or a scalar curl or div to visualize the different fields.

Attachments (2)

Download all attachments as: .zip