Image detection – how, what, where – any insights?

I don’t have much experience with image detection in a video, and wanted to understand what options are possible – all I know it is not as simple as “CSI” makes it out to be – yet. Smile

Here is the scenario:

Say I have a remote controlled aircraft (a mini helicopter) which among other things has a camera fitted which is filming over a certain area. I use a training set and somehow “train” the camera to look for certain objects and recognise data points of that object and learn it to recognize that object (for example a human in Pink). Now once this is trained, if the helicopter is over another area and recording I want it to be able to recognise these pre-trained images in that video feed and do something if something is found (e.g. a human in pink).

The closest “CSI” analogy I can think of is – the facial recognition which they show to be iterating through hundreds of photos comparing data points (which of course as we all know this is mostly fiction – but that is a discussion point for another day).

So the question:

What does it take to compare objects in video feeds? I am not really interested in knowing how to “train” and find a pink human – but rather how to reliably and possibly quickly (within reason of course) compare and see if two are the same of the pattern. This should also reduce the false positives – e.g. a pink table might look like a human in pink with the environmental factors (shadows, sunlight, rain, angle of camera, etc.). I have read a little online but don’t have any experience in the subject.

Has anyone done this (or something similar) in the past or have any pointers for me? Smile