Interpretation of animal behaviours using novel video processing techniques in Artificial Intelligence

点击翻译按钮获取中文摘要

Although some progress has been made in the interpretation of human behaviours using machine learning, research into the interpretation of livestock behaviours using video footage has been more limited. This is unfortunate, as automated systems for herd monitoring could produce tangible improvements in animal welfare. What progress has been made in the development of precision livestock farming has, typically, focused on the use of physical tags or electronic devices such as accelerometers to measure and record data. Although informative, these techniques can be physically invasive or costly. There is, therefore, a need for accurate systems which can characterise and record animal behaviour using already existing cameras, as this would allow wider adoption. The goal of this project is, therefore, to develop novel techniques in machine learning and machine vision systems which will allow for automatic interpretation and recording of livestock behaviours.Data for this project will be provided by an industry partner in the form of raw or annotated video footage of dairy cows. This footage will be used to train novel machine learning systems to characterise and record important behaviours such as standing, walking, eating, drinking etc., and possibly monitor for longer term behavioural changes. Data about these behaviours could be used as an indicator of both overall herd health and the health of individual animals. Experimentation with machine learning architectures, data augmentation techniques, and data pre-processing will be necessary to produce viable solutions.