Plankton 3D tracking: the importance of camera calibration in stereo computer vision systems

Computer vision applications are very useful to study animal movements, but due to their intrinsic complexity they are challenging to design, implement, and use properly. We here describe a calibration procedure for a laboratory stereo vision system for tracking zooplankton in three dimensions (3D). We demonstrate the crucial importance of using a correct calibration for proper interpretation of animal swimming tracks. We also address the effect of the air-water shift phase in the calibration procedure. Actually, not performing a proper calibration caused an average positional error of more than 25 body-lengths in the investigated animal Daphnia magna. Furthermore, we evaluate the different outcomes of using 2D and 3D tracks obtained with a calibrated stereo vision system and show that although 2D tracking might be sufficient in some cases, the method was in our study unable to give information of swimming path geometry and underestimated the speed by 25%. Finally, we discuss consequences for biologically relevant questions when an incorrect methodology is used and strongly recommend that future studies provide detailed descriptions of the framework used for calibration to allow for comparisons between different studies.