Object Detection - finding animal features within an aerial photograph

Question

I have a small team of developers with the task of identifying farm animals in a series of aerial photographs (from a small UAV) of farmland and/or river settings.

I understand this is an open-ended question but can anyone suggest starting places to look for techniques/packages or any other advice for attempting this kind of project? Either image detection and/or machine learning for this purpose..

Any advice is appreciated.

Answer 1

This is actually a two part problem:

1. Separating 'objects' from background
2. Recognizing 'objects' as animals

If you separate this problem into two problems, you can start building for problem 1 and figure out number two on the way.

Problem one is a typical detection problem. Obviously it depends on the images you have, the resolution those images have and the size of the objects relative to the whole image. But this should certainly be do-able. It's a fairly basic Computer Vision problem, depending on your exact situation of course. Search for blocks of pixels that are significantly different from their environment. Then separate those blocks, to use them as input in problem number two.

Number two is actually also really dependant on the dependencies of problem one, however these factors are much more influential in solving for two. It might very well be impossible to properly separate actual animals from say, a water container standing in the same field of grass. But I'd try and solve this in a machine learning way, probably with a neural network. But you'd have to solve 1 first, and have a large enough result set to train with. First separate the animals from the non-animals yourself, and then train the network.

Not really a complete answer, but as you said, it's an open-ended question. Too many unknowns at this point.