Thanks Hans,
Yep, I've been working my way through the 'help' patches. I'm currently exploring pix_multiblob but it's not tracking as I thought it would. When it has an 'argument' to track 2 (or more) 'blobs', it still only tracks 1 'blob'.
Does anyone know if it's 'argument' value only relates to the amount of representations that can be applied to 1 'blob' being tracked?
Thanks, mz_multiblob ^^
Hi, Eleanor
- make sure to use the latest pix_multiblob from recent Gem Git as the
older version doensn't work properly for blobs greater than 10.000 pixels (eg, 100x100).
- not sure I understood your question, actually. isn't that 'argument' the
maximum number of blobs? also, the minimum blob area can be set and is a very useful filter.
- note that detection of a blob, which is what pix_multiblob does, is not
the same as tracking. Traking inolves detecting consistent blob id's across frames, which is not done in pix_multiblob. Ways to do tracking would involve combining pix_multiblob with color detection (like the new pix_colorclassify), the color being the consistent id itself, or just being used as part of a blob similarity function based on things such as such as color histograms (pix_opencv_meanshift would be a standard tracking approach along those lines).
hope this helps,
On Tuesday, November 29, 2011, Eleanor Stewart eleanorstew@gmail.com wrote:
Thanks Hans,
Yep, I've been working my way through the 'help' patches. I'm currently exploring pix_multiblob but it's not tracking as I thought
it would.
When it has an 'argument' to track 2 (or more) 'blobs', it still only
tracks 1 'blob'.
Does anyone know if it's 'argument' value only relates to the amount of
representations that can be applied to 1 'blob' being tracked?
Thanks, mz_multiblob ^^
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 2011-11-30 04:14, Ricardo Fabbri wrote:
Hi, Eleanor
[...]
- not sure I understood your question, actually. isn't that 'argument' the
maximum number of blobs? also, the minimum blob area can be set and is a
indeed. so if you say you want to track "10" blobs, and there is only 1 blob in your image, [pix_multiblob] will still only output a single blob.
i think the main problem people are having with [pix_multiblob] is that it outputs the detected blobs as a "matrix", as used by the "iemmatrix" library.
attached is a small abstraction that will convert the output of [pix_multiblob] to lists, which might be easier to handle for you. (actually there are 2 abstractions, the one in iemmatrix/ depends on the iemmatrix library, the other one is pure vanilla; they should do the same but the latter might be a lot slower with larger matrices)
- note that detection of a blob, which is what pix_multiblob does, is not
the same as tracking. Traking inolves detecting consistent blob id's across frames, which is not done in pix_multiblob. Ways to do tracking would
there is also a simplistic tracker example that comes with Gem (and again depends on iemmatrix) that uses a simple algorithm for labelling blobs, based on the assumption that blobs move as little as possible: it's called [pix_blobtracker]
fmadr IOhannes
-
Eleanor Stewart -
IOhannes m zmoelnig -
Ricardo Fabbri