This is probably one of the outcomes of my work that I'm more proud and excited about! One day my boss said we had to come up with a new idea for a grant proposal, it had to be something innovative and technological. At the time there had been an outburst of articles on Precision Livestock Farming, and how animals and production quality could be monitored and managed in an automatic way, simplifying farmers' work and ensuring high welfare standards for the animals. Then, I came across an interesting article by prof. Cucchiara from UNIMIORE explaining how surveillance cameras in public places could detect in real time people acting differently from the rest of the crowd (e.g. jumping over the turnstile in a metro station instead of walking through it) and send an alarm call to security. That's when my light bulb turned on: automatic surveillance systems to detect early signs of abnormal behaviours in shelter dogs to improve their quality of life! I contacted prof Cucchiara, asked her if my idea could be developed and she said YES! For the following two years her collaborator Simone Calderara and the ImageLab team developed a new prototype software now called B.A.R.K. (Behavioural Automatic Recording in Kennels).
B.A.R.K. is able to automatically infer the behaviour of dogs housed in kennels from 3D visual data and through structured machine learning frameworks.
Example of how the B.A.R.K software detects the main body-parts of the dog, through complex algorithms creates a skeleton and detects movements and variations of postures in time
In its current state the software is able to automatically detect the dog’s posture and activity level inside the pen, without pre-determined schemes and independent from human supervision. This 3D framework was designed to be invariant to the dog’s shape and size and could be extended to farm, laboratory and zoo quadrupeds in artificial housing.
Examples of body-parts and posture detection by B.A.R.K.
The ability to monitor the animals automatically and in continuum allows a large volume of data storage and analysis with minimal effort. Long-term monitoring could allow the identification of predictors of undesirable behaviours such as stereotypies or intense aggressive reactions. A system recording and analysing data continuously can detect rare or unusual patterns of behaviour that could be missed when applying common sampling techniques. Lack of rest, signs of apathy and depression (predicted by unusually low activity) may be signs that the dog is struggling to adapt to the kennel environment and these could be detected through this system in an automatic way. The use of only one 3D camera for data acquisition, makes its application affordable, and can be adapted to almost any experimental or clinical setting. Veterinarians could benefit from a smart monitoring system of hospitalised animals or for the assessment of animals at home with in-real-time data analysis of their behaviour. The study proposed is 100% non-invasive for the animals as it relies on image acquisition of dogs in their home pen with no disruption of their daily routine. To read the open access paper click Barnard et al. PLoS ONE 11(7): e0158748