Digital Signal Processing Reference
In-Depth Information
Frame 834
Frame 909
Frame 945
Frame 979
Frame 1000
Input
Ground
truth
GMM
T2-
FGMM
Our me-
thod
Fig. 2.
The representative results of “canoe”
FP
FN
FP
FN
T2FMRF
T2FMRF
T2FGMM
T2FGMM
GMM
GMM
0
2
4
6
8
10
x 10
6
0
1
2
3
4
5
6
x 10
6
Total Errors
Total Errors
Fig. 3.
The 1
st
column is the total errors of the sequence “overpass”, 2
nd
column is the total
errors of the sequence “canoe”
6
Conclusion
We have developed a new method for motion detection in dynamic backgrounds
based on T2-FGMM and MRF. The proposed method achieves superior performance
than GMM and T2-FGMM, and it is different to other approaches based on Bayesian
method since the fuzzy model is used, which aims to handle the problem of dynamic
backgrounds. Furthermore, the proposed method can be extended to detect shadow
easily. The future work is to develop a more effective likelihood function of the
background and the foreground to make the method more robust.
Acknowledgments.
This work is supported in part by Specialized Research Fund for
the Doctoral Program of Higher Education of China under Grant (20090031110035),
National Natural Science Foundation of China (61203333), and the Fundamental
Research Funds for the Central Universities.
