Monitoring Nesting Seabirds using Computer Vision

Seabird populations are a valuable and accessible indicator of marine health: population changes have been linked with fish stock levels, climate change, and pollution. However, the most recent reports indicate that many UK seabird colonies are in decline, and suffering from low breeding rates. The mechanisms causing these declines are not well understood, though it has been suggested that low breeding returns in a population of Common Guillemots may be related to increases in attacks on chicks while parents spend more time foraging for food. Identification and interpretation of these types of behavioural artefacts are a key factor in understanding the development of the ecosystems in which these populations live, and securing the future health and sustainability of coastal marine environments. This project will pilot the development of computer vision techniques which support automated surveillance of nesting seabirds, and collect behavioural data on a scale not currently available to ecology researchers.The Computational Ecology and Environmental Science group (CEES) at Microsoft Research, Cambridge, UK are currently working with ornithologists from the Evolutionary Biology and Behavioural Ecology group at Sheffield University on a programme to monitor a population of Common Guillemots on Skomer Island (west Wales). Currently, manual inspection is used to estimate the size of the population. However, this process is highly laborious, and it is not feasible to gather more detailed data about individual birds. This places severe limitations on further analysis of this population. This project will be conducted in collaboration with CEES, and will develop computer vision algorithms which analyse video data of a Guillemot cliff nesting area on Skomer and automatically determine the amount of time that parent birds are spending at their nests. It is not feasible to extract this data using manual methods, so this new data will allow CEES to make investigations into the relationship between chick survival and nest attendance. There is little existing work which uses computer vision to monitor wildlife, and so this project will engage with the specific technical problems of automated visual surveillance and image processing in natural environments. Whilst the project is focussed primarily on the Skomer Guillemots, the proposed techniques could be readily deployed for monitoring other seabird species, and also used to support other applications of computer vision.