Comparing fields obtained with different processing parameters are particularly useful to test the precision of PIV measurements.
Differences between two fields inside the same file can be obtained by simultaneously selecting two check boxes on the left of the 'uvmat' interface. For instance let us compare 'civ2' and 'filter2', the filtered field, using the action 'VIEW_FIELD'. First select the field option 'civ2' option alone, remove the scale option 'auto', to fix the arrow scale, then select the field option 'filter2' alone, to see this second field: the same flow features are visible. Then select simultaneously the two options 'civ2' and 'filter2' , and the difference between the two fields is visible. By decreasing the scale 'max' by a factor 10 , one can clearly see some differences, less than 10 % of the 'civ2' velocity. Once the two fields have been selected, all the processing, including statistics and histograms, is performed on the field difference.
The fields obtained in different files can be similarly compared by opening the first field with the browser, for instance, the field obtained from the pair -12|12 from the previous civ3 processing (accessible in 'UV_DEMO/EX_1/VEL_12'). Then press '-' on the upper left near the file name: the name is copied in the window below. Then open the new field with the browser as usual, for instance a pair -14|14 (accessible in 'UV_DEMO/EX_1/VEL_14'). Then any processing is performed on the field difference. The file number in the lower window is automatically varied in coincidence with the first one.
Let us for instance run the 'ACTION' 'time_series' on this field difference, keeping the same domain as before. We observe that the r.m.s remains rather small, of the order of 0.1 pixel. Since the errors from these different image pairs can be reasonably considered as independent, this is a good estimate of the typical measurement error. However the min and max of this velocity difference reach values of the order of one pixel, which means that a few vectors are less precise. The histogram, shown in Fig. 8 (black curve) indeed indicates tails with a small probability of high excursions from the r.m.s. (they are far from Gaussian).
It is now possible to eliminate these erratic vectors by the running the 'fix_vel' ACTION, still keeping the same selection of input files. Then a threshold is asked for the velocity difference, let us choose 0.5 pixel: it will remove any vector whose difference with the corresponding value in -12|12, linearly interpolated at the vector position, exceeds 0.5 pixel. Indeed running again the 'time_series' ACTION, we observe that the min and max remain smaller than 0.5 pixel (Fig. 8 left). The proportion of vectors removed by this operation ('pfalse') is less than 0.1 %, so the field coverage is still excellent, as it can be checked by a direct visualization of the fields -14|14 alone. For that purpose, unselect the '-' check box, and switch to the 'VIEW-FIELD' ACTION. Use the 'see_false' check box on the upper right to see the removed vectors in magenta, or to erase them form the visualization. Note that this removal is reversible: it is possible to come back to the 'fix_vel' ACTION and use a different threshold.
We can state that all the velocity vectors in this final series -14|14 have a precision better than +-0.5 pixel, i.e. 0.15 cm/s (dividing by dt=1.12 s and the scale 3.1 px/cm). This is 5 % of the extremal velocity 3 cm/s. The r.m.s error is however less than 1% (0.1 px).
|