Background Subtraction is a technique in which background and foreground are segmented so that we can perform our required algorithms (such as face detection, gender classification etc). The basic approach for background subtraction is to store the background image as the reference image, in which there is no movement and then in every other frame subtract the reference image to extract the alien objects in the scene. The alien objects are captured as foreground and these are used for further applications.
Since the background does not always remain constant (mainly because of light changes or continuous movement of leaves) therefore we need to develop an approach in which we update the background continuously so that minute changes in the background are made part of the background. Stauffer & Grimson’s algorithm is one of the most reliable algorithms to facilitate adaptive background subtraction.
Stauffer & Grimson Algorithm
The algorithm models each pixel with a mixture of Gaussians. At every frame, for each pixel, distance of pixel’s color value is calculated from each of the associated K Gaussian distributions. Every new pixel is checked against all existing distributions. The match is the distribution with Mahalanobis distance less than a threshold. The mean and variance of unmatched distributions remain unchanged. The matched distributions are updated by using alpha blending concept. For the unmatched pixel, the lowest weight Gaussian is replaced by the new Gaussian with mean at the new pixel and an initial estimate of covariance matrix. Then we sum up distributions less than some threshold to decide whether the current pixel is part of the background or foreground.
Reference
"Adaptive Background mixture models for real-time tracking" by Chris Stauffer and W.E.L Grimson (The Artificial Intelligence Laboratory, Massachusetts Institute of Technology)
http://www.ai.mit.edu/projects/vsam/
Overview of the algorithm is given at the following link:
http://www-staff.it.uts.edu.au/~massimo/BackgroundSubtractionReview-Piccardi.pdf
The original paper is available at the following link:
http://www.ai.mit.edu/projects/vsam/Publications/stauffer_cvpr98_track.pdf
Since the background does not always remain constant (mainly because of light changes or continuous movement of leaves) therefore we need to develop an approach in which we update the background continuously so that minute changes in the background are made part of the background. Stauffer & Grimson’s algorithm is one of the most reliable algorithms to facilitate adaptive background subtraction.
Stauffer & Grimson Algorithm
The algorithm models each pixel with a mixture of Gaussians. At every frame, for each pixel, distance of pixel’s color value is calculated from each of the associated K Gaussian distributions. Every new pixel is checked against all existing distributions. The match is the distribution with Mahalanobis distance less than a threshold. The mean and variance of unmatched distributions remain unchanged. The matched distributions are updated by using alpha blending concept. For the unmatched pixel, the lowest weight Gaussian is replaced by the new Gaussian with mean at the new pixel and an initial estimate of covariance matrix. Then we sum up distributions less than some threshold to decide whether the current pixel is part of the background or foreground.
Reference
"Adaptive Background mixture models for real-time tracking" by Chris Stauffer and W.E.L Grimson (The Artificial Intelligence Laboratory, Massachusetts Institute of Technology)
http://www.ai.mit.edu/projects/vsam/
Overview of the algorithm is given at the following link:
http://www-staff.it.uts.edu.au/~massimo/BackgroundSubtractionReview-Piccardi.pdf
The original paper is available at the following link:
http://www.ai.mit.edu/projects/vsam/Publications/stauffer_cvpr98_track.pdf
Bunty is this English ! I do not understand most of it :p
ReplyDelete