The sheer volume of data collected each year (over 1 million images) presents several challenges when it needs to be accessed and analyzed. Needless to say that automated forms of processing are highly desirable. For this purpose, we have established a close collaboration with Gianfranco Doretto, a colleague in WVU’s Department of Computer Science and Electrical Engineering to develop new computer vision and image analysis techniques that support our cause.
In the past few years, the basic tools for object recognition and facial recognition from visual data have been applied to some wildlife camera trapping problems – for work on polar bears, two types of sharks and other species. However, in each case these analyses to identify individuals required human-aided pre-processing to sort pictures of interest. Our collaborative effort is geared towards developing generalizable image analysis tools for camera trapping data that do not require human-aided pre-processing. Thus, our image analysis will be a two-step process. The first step, similar to object recognition, requires species recognition (e.g., training the software to tell a bobcat from a bald eagle) and the second step, similar to facial recognition, requires individual identification (e.g., training the software to tell golden eagle #1 from golden eagle #2).
For more information on the computer vision research conducted by Doretto’s team and on the individuals involved, please visit the web page of the WVU Vision Lab.
Our work in this area is in its infancy – please check back regularly for updates on this project.