Audio is streamed directly from the publisher (video.tedomum.net) as published in their RSS feed. Play Podcasts does not host this file. Rights-holders can request removal through the copyright & takedown page.
Show Notes
This video is three stacked HD frames representing from top to bottom: the final image, the detected foreground areas, and the detected foreground motion (with normalized intensity).
This is an algorithm I'm experimenting with for making time-lapse videos of these critters and other scenes where the motion content may vary widely from time to time. The basic idea is to detect motion and try to average input frames together so long as the amount of motion is under a threshold. By itself though this algorithm would be far too sensitive to camera and compression noise in the input, so I'm combining it with a statistical model that segments the foreground and background. Implemented with OpenCV and CUDA; I'll be talking about this more and sharing the code after a bit more work.