New facial recognition software designed for study of wild apes
By Ben Coxworth
August 3, 2011
When studying wild animals such as gorillas and chimpanzees, it's not uncommon to use photo or video traps - unmanned cameras that are triggered to capture images when creatures pass in front of them. Scientists can then retrieve the cameras and review the footage, to get an estimate of the numbers of a certain species within a given area, and to see what those animals have been up to. One of the problems with this approach, however, is that it's often hard to tell one animal from another - are you looking at several shots of several different apes, or is it the same individual every time? German scientists are developing wild primate-devoted facial recognition software, in order to answer such questions.
The system is being created through the SAISBECO (Semi-Automated Audiovisual Species and Individual Identification System for Behavioral Ecological Research and Conservation) project, which is a joint effort of the Fraunhofer Institutes for Integrated Circuits IIS and Digital Media Technology, and the Max Planck Institute for Evolutionary Anthropology.
It starts by filtering the stills and footage for images in which whole ape faces can be seen. An algorithm then analyzes the features of each face, and matches up specific faces with specific animals. The system can also pick out ape vocalizations from video footage soundtracks.
When tested on a group of 24 chimpanzees at the Leipzig Zoo, the software managed a recognition rate of 83 percent.
In the wild, however, factors such as poor lighting and foreground obstacles come into play, leading to less of the ideal full-face shots. Under such conditions, the recognition rate drops to 60 percent. In order to boost that rate back up, the researchers are working on more algorithms, that could identify specific apes based solely on individual biometric features such as the eyes, nose and mouth.
The scientists now plan on testing the system with a larger data set, and adding a module that recognizes ape behaviors. Ultimately, the SAISBECO project hopes to use camera traps outfitted with the software to study chimpanzees and gorillas in Gabon's Loango National Park.