Bird migration is not just an outstanding event but is a critical indicator of environmental health, biodiversity, and climate change. For these reasons and so many other reasons scientist are interested in monitoring bird migration and they use a variety of methods for this purpose, including satellite tracking, weather radar, moon-watching, or attaching geolocators on captured birds. All said, if we need a comprehensive and reliable bird monitoring system, we need to detect migrating birds at night which is a much more challenging task to do.
A new paper published in CVPR 2016 proposes a vision-based method for monitoring bird migration during night. By setting up stereo cameras facing up to the night sky, they detect and track migrating birds illuminated from below by light pollution in the recorded videos. They use near-infrared cameras and are able to detect multiple flying birds and estimate their orientations, speeds, and altitudes, which was not apparently possible in previous methods.
However there are three main challenges in using a video based approach. First the videos that are recorded at night inevitably contain substantial noise because of the low-light environment, so the signal to noise ratio is low and that makes it hard to detect birds.
Second, migrating birds fly at different altitudes ranging from several hundred feet to few miles. In this case, the bird may span only 1-2 pixels in a frame. So they had to use motion as a reliable cue for detecting a bird, which again has its own difficulties due to low signal to noise ratio.
And finally, the third challenge would be the need for efficient algorithms for large-scale deployment of such system.
Bird migration is a regular seasonal, large-scale, often long-distance movement between breeding and wintering grounds and if you are into bird watching or have some other related interests in this topic, you should take a look at their technical paper.
If you are a regular visitor of Computer Vision Online, you may have noticed that I'm obsessed with birds (see our other blog posts about computer vision and birds here and here). Also, I'd like to share with you this video of a wonderful flock of birds. Enjoy!