Starting anything is always difficult, however this process is probably not aided by constant procrastination. However when starting research, it is essentially to get your ideas in order, otherwise you end up writing a 15,000 word dissertation in under a week – not something I’d recommend.
The topic of research is one that is out-there and bizarre, and will take across the globe, to census flamingos. Currently, as an ancient Computer Scientist currently in a Geography department I’m still trying to get a handle on what Geography is, currently the definition is “Everything to do with space, attributes and the relations between them”. When I was at school that was topography, which depending on your school of thought has a toe in Geography and a bloody big foot in the Mathematical realm, but that’s just my subjective belief.
So, censusing Flamingos. I know vaguely what I need to do. Firstly get a mobile device on which to do the processing and capturing. This decision has already been made, and it’s the Android Pulse, it’s cheap and does exactly what it says on the tin, which makes it perfect for Lake Bogoria, Kenya, where flamingos live. Secondly develop an algorithm that can detect flamingos. Again this shouldn’t be a problem, many algorithms exist, and their usage/implementation is even documented for the intended platform. Win.
The tricky part however will be the physical counting of the flamingos, if taken from a point of altitude, when the image detection algorithm finds the Gaussian Kernel difference of the picture and by extension gives the outline of the flamingos. From here, how can we count the images, I would think that working out the centre of each distinct shape would give a good result, however flamingos like to bunch up, how would this work then? I would think it is possible to get an average size of flamingo, and when faced with a bunch of them calculate their density.
The problem with all of these methods is that as it’s with a mobile device, the user and camera is going to be at ground level, which means a difficultly in accurately censusing them due to the possibility that on eye level the flamingos are going to be shrouded and hidden by those at the front. I severely doubt there is a method to get past this, which means there has to be some sort of elevation with the user looking down on the flamingos.
With all this in mind, the other chestnut to crack is the issue of real-time workings, taking a picture is all well and good however the ideal solution – in my opinion – would be a real time system that would recognise that you are starting at a defined point, then while panning across the flamingo population ~20,000 of them, counts them as you go, getting all your birds in one sweep so to speak. These are the issues that are going to be keeping me up until September, so I best get started!