Kevin McGarigal has been awarded a $750,000 NSF grant to develop a system of tracking wildlife using digital photography
Posted: January 28th, 2008
Excerpted from UMass News Office
AMHERST, Mass. – The ability to identify and monitor individual animals in their environment can be critical for studying things like migration patterns or habitat use, but traditional tracking devices can be invasive and some populations are just too large to track easily. Now University of Massachusetts Amherst scientists are developing computer vision technology that will allow for quick identification of individual animals using digital photographs.
Much like facial pattern-recognition programs used by the FBI, the software will use complex algorithms to identify each individual’s unique features and automatically catalog them. The work could provide a valuable tool for managing endangered species and for basic ecological research.
The National Science Foundation has awarded UMass Amherst’s Kevin McGarigal $750,000 to develop the technology with former UMass Amherst graduate student Sai Ravela, who is now a research scientist at the Massachusetts Institute of Technology. McGarigal discovered the need for this technology through his study of threatened marbled salamanders, which rely on vernal pools to lay their eggs and breed. His research focuses on their population dynamics-do the salamanders maintain separate, local populations or interact in clusters of interdependent populations? The answer has important implications for conservation efforts; if the populations are separate then conserving individual vernal pools and their uplands is adequate. But if the salamanders depend on periodic interaction with populations from nearby pools then larger areas of connected pools will need to be protected.
To investigate, McGarigal and doctoral student Lloyd Gamble tracked thousands of individual salamanders on their way in and out of vernal pools. The research team had already built a system to take digital photographs of each individual salamander passing through and had archived several years of photographic data. But with hundreds of photographs being taken each night, it was almost impossible for his team to identify and catalog each individual. So McGarigal turned to computer vision technology. McGarigal, Gamble and Ravela then developed tailored algorithms that look at unique salamander markings and identify individual salamanders. The technology enabled them to translate thousands of individual photographs into a gold mine of data. For the first time the researchers could document how often individual marbled salamanders tried to breed, how long they survive as adults, and how often they return to the same pool or disperse to a new one. This new information allowed them to create the first models of the dynamics of marbled salamander populations. The technology may be especially useful to biologists who are trying to track thousands of individuals within a single population, an effort that has been difficult in the past due to the flood of data. "Our technique lends itself well to large biological databases where there is a real efficiency need," says McGarigal. The new technique also eliminates the need for invasive procedures, such as physically attaching a marker to the exterior of the organism or surgically implanting one in the body cavity, practices that can potentially alter the normal habits of the individuals, says McGarigal.
After publishing their initial results, McGarigal and Ravela heard from many colleagues, wondering if the same technology could help them with their wildlife research projects. The inquiries inspired them to seek funding from NSF to investigate whether the technology could be transferred. Now the scientists are tailoring the algorithms to other amphibians such as frogs and toads. They hope that eventually they can apply the technique to a wide range of patterned animals, such as elephants or leopards. "Now we can test our current algorithms and expand them to deal with a broader array of individual pattern recognition problems in the field," says McGarigal. "We’re excited to see how far we can push the envelope in wildlife recognition with this new technology."