Towards a holistic assessment of animal welfare

It is important to understand how animals feel, as this is an essential part of animal welfare. People can recognise emotions by observing facial expressions, body language and movements. This holistic approach has also been used to study animals. Therefore, the movement of several body parts can be combined to understand, if an animal is moving in a relaxed way (e.g. playful jumps) or anxiously (e.g. tense movements of legs, widened pupils, erect ears). Based on this, the scientific technique called Qualitative Behavioural Assessment (QBA) is a method where humans observers are asked to describe with their own words and score emotional states, that animals express when they engage with their environment. But we still don`t know which specific features about how an animal looks or moves can tell us about its feelings. Also, the behavioural observations mentioned take a lot of time, so they cannot be used in many commercial situations, e.g. on farms. Artificial intelligence is already used for recognising human behaviours. Therefore, our project aims to explore this new way of using machine learning models (through computer vision techniques) to understand how pigs are feeling by looking at their body language using videos. These models will be trained using many examples of videos from individual pig body language when experiencing different emotional states, for example, positive emotions during feeding or negative emotions when in an unfamiliar environment. As a result, we should be able to link how an animal moves with how it feels. As part of developing the model, we will use eye tracking during experiments to explore how people understand animal body language. The model will be used and tested in different situations on commercial and experimental pig farms, with groups of animals of different ages (piglets, growing pigs, sows) and in different housing conditions (conventional, straw, free range). Automated detection of body postures and movements means that we can monitor more animals for a longer time. This system could help to monitor the welfare of animals (for example, on farms or at abattoirs) and learn more about individual animals in experiments.