1. Classic methods: fast but work only under speciffic conditions: average, median, etc.
2. Pametric methods: Mixture of Gaussians[3]
3. Non parametric methods: Kernel Density Estimators[2]
4. Mean Shift
5. etc.
Automatic Background Substraction
Many computer vision applications require an estimate of the background and foreground of a given image. For example tracking algorithms[1] that try to estimate the 3D position of a subject using only videosequences. Many different types of background substraction exist:
|
1. Classic methods: fast but work only under speciffic conditions: average, median, etc. 2. Pametric methods: Mixture of Gaussians[3] 3. Non parametric methods: Kernel Density Estimators[2] 4. Mean Shift 5. etc. |
Figure 1: background substraction of a golfer.
1. Classic methods: fast but work only under speciffic conditions: average, median, etc.
2. Pametric methods: Mixture of Gaussians[3]
3. Non parametric methods: Kernel Density Estimators[2].
4. Mean Shift
5. etc.
For a review of most common techniques see...
In this projet the student will first study the different types of background substraction techniques in terms of the following criteriums: speed, memory and accuracy, and implement the most relevant ones.
Reading:
[1] R. Urtasun and P. Fua, "3D Human Body Tracking using Deterministic Motion Models". European Conference in Computer Vision. Prague 2004.
[2] A. Elgammal, D. Harwood and L. S. Davis. "Non parametric Model for Background Substraction". 6th European Conference on Computer Vision. Dublin, Ireland, June/July 2000.
[3] C. Stauffer, W. E. L. Grimson. "Adaptative background mixture models for real-time tracking". Proc. of CVPR 1999, pp. 246-252.