Wildlife Tech Class of 2014
Mamaroneck High School
Project: Using Automatic Species Recognition Software to Identify Animals in our Wildlife Camera
Project Status: Completed
Camera traps have been used as an effective way to sample wildlife. However, biologists often have to sort through a large number of pictures to find a few photos of their target species. The open source program “Stripespotter” is an automatic individual animal identification system for animals with prominent patterns. I examined camera trap photos from four New York Parks and evaluated Stripespotter’s ability to discriminate between coyotes (Canis latrans), raccoons (Procyon lator), foxes (Vulpes vulpes), and empty background pictures. I examined 296 pictures taken in Riverdale, Inwood, Alley Pond, and Pelham Bay Parks. Stripspotter was “trained” with 21 coyote, 5 raccoon, 10 fox, and 8 empty background pictures. I then examined 296 different pictures taken at the same locations to see if Stripespotter would correctly identify photos with or without the target species. I conducted four different statistical analyses to determine the accuracy of each species. Stripespotter was able to correctly identify 70% of the pictures. It offered visual feedback to a human user, essential as part of a computer-assisted system. While not yet fully successful, the results suggest that with additional training and development, this type of application holds potential for facilitating the work of wildlife biologists in a variety of contexts.