06. Tracking Recommendations

Three simple recommendations for successful tracking

In this part we would like to give some simple recommendations on how to plan and execute experiments that produce trackable videos.

Rule 1

First of all, if the subjects you want to track with BioTracker (or any other tracking software) are not sufficiently different in coloration from the background, a proper tracking will most likely fail. As an example, a white mouse recorded from above in a white-bottomed cage is a bad experimental choice. Most tracking algorithms use some kind of background subtraction, which basically requires that the subject's pixels are distinguishable from the background pixels. Also, tracking will be hard if the subjects were not moving. In this case, many algorithms that compare frame by frame which is the background (stable) and which is the subject (unstable) will fail to tell apart subject from background even if coloration differs. This also means that a moving, unstable background is bad for proper tracking outcomes. So, we recommend to design experiments in which subjects differ in coloration from the background, subjects move (even slowly) and the background is stable (not moving, changing). In many cases, experimenters record their whole experiment, which is a good thing and recommended as good scientific practice. However, if the experimenter introduces the animal into the arena, hand or fingers (or dipnets or cages) are visible in the video and thus hamper a proper tracking. We thus recommend to either cut recorded videos to the essential parts - the phase in which the the animal should be tracked - or, depending which tracking module is loaded in BioTracker, to fast forward to those parts and only then initiate tracking.

Rule 2

Second, videos are a bunch of pixels (every digital picture consists of these small color squares) and tracking softwares just look for those pixels but not for cm or any other "real-world" measures. Thus, one has to tell the computer what size a pixel is in the real world. As this measure varies depending on your recording mode (camera-object distance, video resolution as well as recording quality), you have to have something in the video of known size you can use to calibrate pixels to real world measures. In the BST tracking module, a rectangle can be drawn upon points of known distance in the video and thus pixels are converted to real world measures.

We thus recommend to always put some kind of grid of known dimensions in the recording frame. This does not necessarily need to be inside the arena but well visible in the video.

Rule 3

Third, we just want to stress that recording orthogonal to the arena's bottom makes life easier. If you record your subjects with an angle, a simple calibration of pixels to real-world measures is not possible that easily anymore.