Preliminaries – Background Subtraction Greg Mori CMPT888
Outline • A little bit on background subtraction
Background Subtraction
Basic Pipeline Background Mixture of Background Gaussians Image Model Median value EM fitting Likelihood under Absolute value of difference with of each pixel background model, over time background, thresholded thresholded (or not)
Many Pitfalls • Background subtraction is not perfect in practice – Background changes – Foreground looks like background – Objects stop moving – … • But it’s a useful signal, combine into human detector
Summary • For more information: – Chris Stauffer and W.E.L. Grimson. Adaptive background mixture models for real‐time tracking. CVPR 1999 – Kentaro Toyama, John Krumm, Barry Brumitt, and Brian Meyers. Wallflower: Principles and practice of background maintenance. ICCV 1999
Recommend
More recommend