An introduction to collecting behavioural data
In this section, I hope to give the overview I wish I'd had when I started collecting behavioural data. Both in how to turn videos into useful information, and in how we can interpret the patterns we see in that information.
What do we want to look at?
In motion: a hoverfly (Episyrphus balteatus) navigates between flowers.
Movement is a key animal characteristic, and iconic of our group. Images of preserved specimens or fossils can tell us huge amounts about what those animals were adapted to do whilst they were alive. We can measure colours, patterns, dimensions and many other characteristics. However, what we cannot do is concretely know how they moved through the world, nor the subtleties of their behaviour. In order to do this, we're going to need videos. Furthermore, to study this well, we're going to need more than one camera. If we have this, we can track the movements of the animal and reconstruct them in a quantitative manner. This is the fundamental basis of much modern behavioural study (ethology) and the advancements in camera technologies, computing power, and machine learning have greatly increased the scope of how we can think about behavioural data.
Insect flight is fundamentally difficult to study
On a warm day, the air around us is filled with an aerial planktonic soup. This can include wind-blown dust, pollen and seeds, but most of it will be will be arthropods commuting from place to place. You can best see this principle if you look sunwards across a garden or through a wood when the sun is slanted low. Little specks of light will be motoring their way from shrub to tree, by the hundreds or thousands, depending on the location. This phenomenon is around us for much of the time and yet we know almost nothing about it. This is because insect flight is fundamentally extremely difficult to study. There are two simple reasons for this: (1) insects are mostly small nowadays, (2) they travel extremely fast, at least for their size. The latter is a consequence of scaling limitations. A 5th Generation fighter jet, travelling twice the speed of sound might just make it to travelling 35 of it's bodylengths per second. In contrast, the average housefly has a top speed near 4 m/s, pushing near 700 bodylengths a second. This is a problem. With modern optics we can amply zoom in to insect scales, but we cannot then track that field of view at the requisite rate to keep up with the animal. Think of trying to use a long-zoom camera lens to track a ping-pong ball during a match and you begin to get the lack of picture. Small size also means that at wider angles of view, it can be hard to even resolve where the animal is against complex natural backgrounds.
Standard Camera Tracking
Lab Based:
The nice bit about being in the lab is that we have control over the conditions. The worst bit about being the lab is that most animals hate being there. For our purposes, tracking free-flying animals, we can now control the lighting conditions and surfaces to help pick animals out from the background. Generally, this means having a pale, plane background against which your subject is neatly contrasted against in each frame. We can use this contrast to distinguish
