1
$\begingroup$

Hello I'm looking for some advice making a animal Classifier from trap cams, I applied SIFT to detect points of interested but it turns out that most of the points were in the background and less in the animals. My intention is to recognize animals from a dataset, any ideas that you can bring me to achieve what I'm trying to do?

Images are something like this and the keypoints : enter image description here

Any other options or suggestions? to change the way I'm going now.

$\endgroup$
1
1
$\begingroup$

You need to do some Background Subtraction on the images. If you have the Background image without the animal, you can simply subtract it from the current image to get just the animal. Once you have just the animal, you can apply SIFT or CNNs or whatever. This is called frame differencing.

enter image description here

If you don't have the background image, you can try methods like this provided by opencv

Basically what you are looking for is background subtraction/foreground detection.

Hope this helps.

image source: http://docs.opencv.org/3.3.0/d1/dc5/tutorial_background_subtraction.html

$\endgroup$
1
  • $\begingroup$ I suspect the OP's background will be more noisy and varied than the example due to lighting changes, movement of vegetation etc. Hopefully they can set the camera trap to take e.g. hourly photos so that they have a close match for lighting available. $\endgroup$ – Neil Slater Sep 22 '17 at 7:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.