Hi Thomas,

One thing you could do (I know this is quite easy to do in Matlab so I
assume it owuld be in R as well) would be to export a movie with a
timestamp on the images if they aren't already timestamped. Then watch the
movie and note the spots where something happens, then refer back to those
images. This assumes the images are timestampped in the name or you can
pull the capture time info from the jpg exif data. Again, this is pretty
easy to do in other programs so I'm assuming someone has written code to do
this in R or Matlab. It also assumes that "events" are not too common so
manually finding the selected images isn't to onerous.

One issue you want to be aware of is making sure that your cameras have the
time set correctly. Many game cameras just give the images sequential names
as they ar captured (like img_123.jpg, etc), so you have to manually rename
them so the image names have a timestamp in them. you can skip this step
but it can be really hard after the fact to figure out what images you are
dealing with if they don't have timestamped names, particularly if you have
lots of cameras and are juggling SD cards with different sets of images on
them. We wrote a windows app that does this (see below) but I bet you could
write a quick script in R to do it as well.

I agree there is probably existing code for finding motion in the images.
At the very least you could start with a simple algorithm
that subtracts the color values of each image from the previous one and
then sums the total value left over. Then test what threshold
value accurately flags the images you want. You can also use the sum of all
rgb values to eliminate nightime or too dark images etc. But I agree with
the other comment that you probably want to run through them all at least
once to make sure you didn't miss anything.

If you have software like Adobe after effects (or the Mac equivalent), you
can also load all the images as a timeseries (they provide this feature for
people doing timelapses), then just scroll through the movie with the
timebar and note what images have movement in them. Video
editing software also provides options for marking key points
and exporting single images so you could come up with a pretty easy work
flow this way. This would probably be the easiest way to do this if you
aren't going to automate it. I think you can also load pretty large image
sets into quicktime pro and do the same thing.

Email me if you want a copy of our naming utility (for windows). It takes
any number of time-series images, extracts the timestamp from their exif
data and renames them with correctly timestamped names in time-series
folders. It works great but since it is optimized for higher framerate
timelapse and use by our timelapse software it does put everything
into folders by hour which can be annoying if you don't have that many
images per day. You also have the option to stick them all in one folder.

Cheers,

Tim

---
Tim Brown
http://Time-Science.com <http://time-science.com/> - Advanced Ecological
Imaging
http://www.gigavision.net - Gigapixel timelapse systems
[email protected]
(801) 893-1314 | (866) 411-3836
 <http://www.linkedin.com/profile/view?id=13027940>
<http://timescience.wordpress.com/>
  <https://twitter.com/#!/timescience>

Reply via email to