ABSTRACT: For decades, external markers have been crucial to the study of avian population dynamics and ecology. However, many traditional methods can have deleterious effects on performance and welfare, while the population data retrieved is often sparse or comes at a significant time and effort cost. Computer vision can offer new perspectives that minimize disturbance and manual effort. Using specifically designed software, we demonstrate the possibilities of automated animal biometrics as a means to provide non-invasive monitoring of colonial seabirds at Robben Island, South Africa.
We will show that a computer vision system designed to monitor African penguins (Spheniscus demersus) using unique plumage features is capable of robustly matching individuals to a population database under field conditions. False individual identifications occurred in under 0.01% of comparisons, while the monitoring capacity of the system was estimated to be above 13% of birds that passed a camera during a trial period. Theoretical and empirical development of this capacity suggests high levels of enrolment and recapture over time frames of a few months. In addition, a test system for visual nest observations on bank cormorants (Phalacrocorax neglectus) provides an example of how the approach can be generalized to provide information on difficult to access or easily disturbed species, without colonial intrusions.
Finally, we outline the current state-of-the-art, the limitations and the practical preconditions for the use of computer vision systems as tools in non-invasive field-identification in general. To date, the technology provides a viable and timely alternative to complement and extend existing monitoring efforts.