Help Understanding The Valididty of Results

[C Lw]

Hi all,

I've been using PIVLab to study the dynamics of ceramic particles moving in low density air. Without going into too much detail, the air is stationary, and the particles are not; the time resolved particle flow is of interest to me. PIVLab has produced some really excellent results, but I want to make sure I understand the results before I proceed.

(I've been feeding PIVLab binary images to make filtering easier, hence the stark black and white; as a side note, if this is a bad idea, I'd love to know!)

 The top image is the filtered result with the Standard Deviation and Local median filters applied; the middle image is the raw result; the bottom image is what happens if I only allow vectors in 32 px interrogation areas that have some white in them, in addition to the Std and local median filters.

I look at 150 frames of data where the flow is changing over time; If I use the red field, the space/time averaging of the flow field I do becomes really unstable and it is basically unusable for the type of analysis I want to do.

My question is: does the top image look valid?  It seems like PIVLab decided to put vectors in a lot of places that are near particles but not necessarily right on top. Is the top photo just full of spurious vectors, or is there some way the PIVLab is deciding to put vectors there in a rigorous way? I basically do not understand the FFT Window Deformation algorithm...

The settings I used was 3 passes: 128 px, 64 px, 32 px, each witha 50% step size. Interpolation is completely off for this analysis.

I've attached an example of of a raw binary image pair where the system is definitely "vortexing" (600A and 600B), and a raw image pair where it is definitely not vortexing (878A and 878B). I've also attatched the raw images (not binary) images of the vortex motion in case this is helpful

Can I reasonably use the results of the top image, or is it back to the drawing board? And if so, are there better settings I could be using in order to extract the particle dynamics? As an aside the camera I have access too is very cheap and slow, and I don't have a laser for imaging

Thank you so much in advance! I really appreciate both the software and any help I receive!

Best,

Callum

[Marco Polin]

Hi Callum,

personally, I don't see why the first image should be a problem. There is no need why the vectors should be "right on top" of the features... after all your 32x32pxl window is quite large. The only thing is that the pattern within the window should allow the program to get a nice, well defined peak in the cross correlation map. 

Marco

[William Thielicke]

Hi, I would not recommend to binarize the images. PIV expects Gaussian intensity distributions for the most accurate results. Have you tried to subtract the background?