Detecting objects in a photograph with opencv

Two faces being detected using

For a project I am working I need to detect simple rectangular objects in a photograph. After some investigation I have settled on using the Open Computer Vision (opencv) libraries to do this. I have played around with opencv a little in the past but still consider myself a newbie. Having said that it has not been too hard to get up and running.

My programming language of choice is ruby so I am making use of the ruby-opencv gem. Unfortunately this gem doesn’t seem to be very well supported and didn’t install using good old gem install opencv. What I found was that this repository seemed to be the most up to date fork of code. So I cloned it and went about getting it installed.

ruby-opencv is a native gem that wraps the opencv libraries so there is not much ruby going on in there and plenty of c++. After muddling my way through a few different messages about dependencies I did not have and some confusing error messages – which again where simply unmet dependencies I finally managed to get the gem built.

The repository has a sample that shows how to do face detection. It is one of those pieces of code that make you go what the heck! it can’t be that simple! And while you can do face detection in about 8 lines of code it is not really the code that is the magic part here.

The face detection algorithm is actually a set of Haar-like features. This is basically a description of what sort of object we are looking for – but go read the wikipedia article for a more comprehensive description. In order to detect a specific type of object we are going to have to write our own classifier that describes what we are looking for.

The next step is to write a classifier for our objects which we will do in the next post.