Some Approaches to Finding Birds in Video Imagery

Abstract

Manually censusing and monitoring bird colonies can be a time consuming, frustrating, and sometimes error prone process. Automation of censusing would efficiently provide consistent and easily collected data which would aid in the monitoring of bird colonies. This paper analyzes some typical scenes from bird colonies and describes our experiments of applying well-known computer vision techniques to those scenes. After a short description of the template matching, convolution, and motion analysis methods we used, we analyze four different classes of scenes using those techniques. They are as follows: -Kittiwake colony scenes are analyzed by convolution with a filter especially constructed to find light circles with dark edges. Results of that analysis show an improvement over directly thresholding such images. -Murre colony scenes prove to be highly resistant to analysis because of the coloration and crowding behavior of murres. However, we do show that it is possible to identify murres with profile presentations in such scenes by using template matching. -Motion analysis techniques are applied to find auklets loitering on the surface of their talus slope nesting areas. Though the birds are practically invisible in these scenes, we show it is possible to find areas of high movement indicating the presence of birds. -Thresholding and convolution are used to recover counts of birds from scenes of flying auklets. Ideas about tracking individual birds through such scenes are also explored.