Abstract
The range and effectiveness of surveillance systems can be effectively improved by the use of the autonomous pan-tilt cameras, rather than the common static cameras. The presently used approaches for the construction of background models fails here as these are for use with the static cameras. The camera motion parameters are estimated and it is used for the purpose of updating the background model online in the presence of scene activity. The motion of the camera is estimated from the median learned over Flow matrix between consecutive frames. The single Gaussian model is used for the detection of the foreground regions are detected as changes in this model. This is skipped during the process of background model construction.
Active video surveillance involves using several static cameras as this can provide a better view of the scene under surveillance. But this system can be expensive and can be infeasible under certain situations.

Get the full report here:
http://scribddoc/37324497/Bachelor-Thesis-Aneesh-Sharma