Facial-recognition software could help to save great apes
Attempts to save populations of great apes could be helped by new facial-recognition software designed to monitor the animals in the wild.
Researchers from Germany’s Fraunhofer and Max-Planck societies are designing a program that can recognise individual apes from photos, video and audio footage recorded in a specific area and so help to count the numbers living there.
‘The biologists [looking after apes in the wild] have to evaluate whether a management strategy is efficient or not,’ Alexander Loos from the Fraunhofer Institute for Digital Media Technology told The Engineer.
‘They have to know the number of individuals of a specific species, whether a population is declining or increasing and which factors influence the population.’
Remotely operated cameras and audio equipment are already used to help monitor animal populations in the wild, but they often produce more data than can be manually processed.
The new semi-automatic system can filter the footage to find where the apes clearly appear and then identify individuals in real time using complex algorithms, in a similar way to human face-recognition software such as that of the Microsoft Kinect.
‘This technology has to be adapted but the similarities of the human face and the ape’s face are clear and so we decided that it is a good idea to use face-recognition software,’ said Loos.
This will be augmented by audio-recognition software that can classify sounds such as threatening grunts and chest drumming, providing information about the animals’ behaviour, as well as helping to identify individuals.
While the software uses similar principles to that designed to identify human faces, it has the particular challenge of needing to work in poorly lit conditions and without the subject standing still and facing directly into the camera.
A test of the first version of the software using 24 chimpanzees from Leipzig Zoo achieved a recognition rate of 83 per cent, but this dropped to less than 60 per cent in poor light or when the faces were partially obscured.
The software has so far relied on algorithms that look at the features and shapes of ape faces, but the researchers now hope to improve it by adding the ability to detect more individual characteristics such as wrinkles or facial marks.
Loos said it was difficult to set a specific target success rate but added that 90 per cent in captivity and 75–80 per cent in the wild would be good.
The team, which also includes academics from the Fraunhofer Institute for Integrated Circuits and the Max Planck Institute for Evolutionary Anthropology, is one year through the three-year project.
The researchers are also working with a team from Bristol University, which has developed software to identify penguins according to their body markings.
Other similar research has looked at recognising dolphins from their dorsal fins and elephants based on the shape and condition of their ears.