next up previous contents index
Next: ofdraw Up: Reference Previous: hs_flow   Contents   Index


$ \bigcirc$Name

motionseg Motion Segmentation (Aubert-Deriche-Kornprobst method)

$ \bigcirc$Command Synopsis

motionseg [-n n] [-p prec] [-e e] [-a alphac] [-b alphabr] [-c alphacr] [-s s] fmov1 fmov2 fim

-n n : Maximum number of iterations, default 100

-p prec : Numerical precision, default 0.001

-e e : Parameter of relaxation, default 0.0005

-a alphac : Parameter alphac of the functional, default 1000

-b alphabr : Parameter alphabr of the fuctional, default 10

-c alphacr : Parameter alphacr of the functional, default 10

-s s : threshold, default 0.25

fmov1 : Original Fmovie (input)

fmov2 : Motion segmentation (output Fmovie)

fim : Background estimation (output Fimage)

$ \bigcirc$Function Summary

void motionseg (n , prec , e , alphac , alphabr , alphacr , seu , N , C , B )

int *n ;

float *prec , *e , *alphac , *alphabr , *alphacr , *seu ;

Fmovie N , C ;

Fimage B ;

$ \bigcirc$Description

This module implements an original approach to the problem of motion segmentation. This approach allows to deal simultaneously with the problem of restoration, allowing the motion segmentation to influence the restoration part and vice-versa.

This algorithm implements the minimization problem presented in [ADK99]:

$\displaystyle \inf_{{B,C}}^{}$$\displaystyle \int$($\displaystyle \int_{t}^{}$$\displaystyle \int_{\Omega}^{}$C2(B - N)2 dxdt + $\displaystyle \alpha_{c}^{}$$\displaystyle \int_{t}^{}$$\displaystyle \int_{\Omega}^{}$(C - 1)2 dxdt +

$\displaystyle \alpha_{b}^{r}$$\displaystyle \int_{\Omega}^{}$$\displaystyle \phi_{1}^{}$(|$\displaystyle \nabla$B|) dx + $\displaystyle \alpha_{c}^{r}$$\displaystyle \int_{t}^{}$$\displaystyle \int_{\Omega}^{}$$\displaystyle \phi_{2}^{}$(|$\displaystyle \nabla$C|) dxdt)

where N(x, y, t) is the original noisy sequence for which the background is assumed to be static, $ \phi_{1}^{}$ and $ \phi_{2}^{}$ are convex functions which smooth the image and preserve the discontinuities in intensity and a, b and c are positive constants. Ideally we would like B(x, y) to be the restored background and C(x, y, t) the sequence with indicates the moving regions. We would like C(x, y, t) = 0 if (x, y) belongs to a moving object and 1 otherwise. See [ADK99] for the details on the motivation of the functional and the existence of solution to the problem. The mathematical algorithm for the minimization in an appropriated space is also found in the article.

Exemple of use:

motionseg cmovie /tmp/m /tmp/bg
fview /tmp/bg &
cmview -l /tmp/m &

$ \bigcirc$Version 1.0

Last Modification date : Thu Apr 15 07:07:02 2004

$ \bigcirc$Author

Toni Buades

next up previous contents index
Next: ofdraw Up: Reference Previous: hs_flow   Contents   Index
mw 2004-05-05