Machine Vision Techniques for the Evaluation of Animal Behaviour

Over the last few decades the nature of farming has changed dramatically with small labour intensive farms being replaced by large, highly automated farms. With this change in the way things are done comes a new concern for animal welfare. Traditionally a skilled stock-man would look after a relatively small number of animals and have much direct contact with them on a daily basis. On the modern automated farm a few people will look after several hundred animals and as such there is much less direct contact.

Within this thesis a toolkit of methods is presented which is suitable for the building of automated monitoring systems within a farm environment. Animal shape is modelled by a two-dimensional hierarchical contour model and various models for animal dynamics are evaluated. A framework is proposed that combines these spatial and temporal models as a combined tracker and behaviour analysis system. A method is also presented for the identification of individuals from their characteristic markings which utilises the result from the animal tracker method. The methods presented have wider application than simply the field of animal behaviour monitoring and applicable to other problems areas within machine vision and beyond.