Continuous and automating monitoring of animal behaviour, such as activity, can help improve the quality of animal experiments. Automated measurements may provide additional insights into the effect of the treatment on an animal’s functioning and may support in making decisions on the application of scientific and/or humane endpoints.

Nowadays, a myriad of sensor technologies exists. However, technologies in the context of laboratory experiments often focus on rodents housed individually or in small groups, and technologies that are developed for a commercial livestock setting are often less suitable for an experimental setting. Therefore, we aimed to develop a tool that could be used in various (non-rodent) animal experiments at WUR, initially focusing on infectious disease studies at Wageningen Bioveterinary Research. We set several criteria to this tool. It should be:

• able to monitor behaviour, at least activity, of individual animals in group-housing;

• able to monitor continuously for a period of 8 weeks, preferably without any interventions that require handling of the animals (such as battery recharging);

• able to monitor (near) real-time, to allow for direct support in decision-making;

• applicable to a wide range of (non-rodent) animal species, including at least chickens, sheep, pigs and cattle;

• applicable to an indoor experimental setting with at least biosafety levels 2 & 3, meaning that it should be possible to easily disinfect (i.e., good waterproofness);

• as simple and sustainable as possible, in terms of installation, few (preferably no) dependencies on external parties, affordability, and re-usability across experiments.

We investigated the use of ultra-wideband (UWB), computer vision (CV) and accelerometers in a couple of experiments. Although UWB and CV harbour potential for more detailed behavioural analysis, our experience is that they have limited generalisability (e.g., due to sensor weight with UWB, or extensive training of CV models), are accompanied with substantial practical challenges (e.g., complex hardware setup, large amounts of data produced), and provide less robust time series (e.g., due to flat batteries with UWB, or occlusion in CV). Therefore, we concluded that accelerometers were most promising for our goal.

After exploring three different accelerometers, we opted for the MoveSense HR2 sensor, a BLE edge computing device which – aside from the 3-axis accelerometer -, also has a gyroscope, magnetometer, and heart-rate and non-medical ECG options. We focus on activity, using the “vectorial dynamic body acceleration” (VeDBA) as activity measure. We believe the VeDBA is a useful measure for our purpose due to its ease of use (That is – being rather simple to calculate), and its strong correlation with energy expenditure (e.g., Gleiss et al., 2011). We currently sample the acceleration (x, y and z) at 13 Hz as raw data and compute/transmit the VeDBA per 1 minute. In this way, we can monitor at a detailed level for multiple months without recharging, while we don’t produce huge amounts of data (<0.2 Mb per animal per day, as compared to video files of at least 5 or 10 Gb per day). We also developed a processing tool with interface, providing caretakers and researchers with a real-time overview of individual animal and group level. Given the relatively light weight of the sensor (10g), it can also be used for a wide range of species. The approach is furthermore flexible, since the settings can be customized per experiment, species and/or research question.