Learn how easy it is to sync an existing GitHub or Google Code repo to a SourceForge project! See Demo
I would like to know if JPIV software is able to monitor/measure/evaluate (don't know what the correct verb is) boulder displacements from aerial images ?
in principal, yes, analog to the following, very simple, example:
There are some details to think about, of course. Your images should have been taken from the same perspective, for example. They must be aligned (with align-image-stack for example). The point of view should be perpendicular to the slide pane, otherwise you have to take into account that the scale varies within the image. For the rest, I guess the same rules apply as for PIV with "normal particles", like that your interrogation window in the first pass should be at least twice as large as the maximum displacement and so on.
Success with your experiments!
Thank you !
I am new to using this software and after several unsuccessful attempts, I wonder about several points.
So I used the software to compare two orthophotos in order to measure boulder displacements on ground surface. I got results that did not satisfy me.
The problem is probably the result of bad configuration parameters which lead me to ask you several questions:
- What is a "pass"?
- In which case i have to use one, two, three pass?
- How to determine the correct settings in the "interrogation supply"?
I hope you will have time to answer my questions.
A pass is a repetition of the PIV evaluation. The result of the preceeding pass is used to preshift the interrogation windows of the next pass. Example: The interrogation window may be 32 by 32 pixels large. When the displacement is 10 pixels in y an 10 in x directions, your aligned (correlating) window position overlaps 22 by 22 pixels. In the second pass, you use slightly different interrogation windows: From the first image, you use an area that is shifted 5 pixels forward in both directions, from the second image, you use an area that is shifted 5 pixels backwards in both directions. As a result, you get two interrogation areas that almost perfectly match over the whole area of 32 by 32 pixels. Your interrogation peak will be very high and sharp and almost in the middle of the area (opposed to a more flat and blunt peak shifted ten pixels in x and y direction). Adding the displacement (the peak position) of the second pass to the ten pixel displacement estimated in the first pass, you get a better estimation of your displacement with sub-pixel accuracy.
As long as you are not experienced with PIV, I would just use one single pass. More passes won't change the result dramatically, it is just slightly more accurate. In fact, already the first pass will give you some sub-pixel accuracy. Just using one pass, you can vary the window size parameters faster and study the effect. Doing this, you will quickly learn and understand PIV evaluations.
Here is an example for a fist pass and a second pass correlation function:
The second correlation peak (right image) is higher than the one of the first pass (left image), although a much smaller interrogation window was chosen in the second pass. You can thus also use multi pass evaluations to enhance the spatial resolution of your evaluation.
Also refer to the manual for additional explanations.
This introductory page might also be interesting for you.
Thanks a lot for these explanations !
I practiced on the software yesterday and I begin to understand it better.
Nevertheless, I have another question.
Do you think that using aerial images taken at different hours of the day (which means that shade on the ground is different) can interfere with the result quality/accuracy ?
Sorry for asking questions before reading the documentation.
After reading documentation I understood that i first have to do image preprocessing to remove noises such as solar illumination!
no problem - the structure of the manual pages is anyway somewhat confusing here and there… I would say, you are right in removing shadows of different length and direction, they would certainly bias the result - although I guess that this is not a straightforward task - would a spatial filter on a FFT transformed image work? It might also be possible to calculate the bias based on a theoretical model of the correlation. That could easily be verified/calibrated by means of a static, but differently illuminated, grainy surface. Such a work would be a basis for a useful publication, I guess…
Success with your work!