Found on Youtube this video illustrating an example of CS background subtraction which I think is an implementation of Compressive sensing for background subtraction by Volkan Cevher, Aswin Sankaranarayanan, Marco Duarte, Dikpal Reddy, Richard Baraniuk, and Rama Chellappa. From the accompanying text of the video:
This is funny as it reminded me a little bit the saliency maps in Laurent Itti's work. At the very least it could be used as a first iteration of these saliency maps as shown in this video.Compressive Sensing based background subtraction.This video contains four parts top-left is the original sequence, bottom left is the reconstructed (using l1-minimization) difference image, top-right is the silhouette (estimated from bottom-left by median based thresholding) and bottom-right is the foreground reconstruction.----This was part of course project for Video Processing (Dr. Baoxin Li) in Fall 2008.Team: Raghavendra BhatRita Chattopadhyay
Hey Igor,
ReplyDeleteIt is really interesting to see how the L1 minimization is being used to model the BG in the demoi shown, but i am trying to understand how this is similar to the Saliency map you have suggested. The basis that has to be chose for L1 minimization needs to correspond to some feature of the background ( is this how you mean it ?)