Thus it is not suitable for real time or live video surveillance applications. The other simple approaches such as frame differencing algorithm or background subtraction algorithms are much suitable for real time applications as they are less complex and provide faster results compared to other methods. Background subtraction technique is a most suitable technique to detect stationary foreground objects as it works well when camera movements are stationary as well as changes in ambient lighting are gradual. Hence it is a most popular technique to separate out foreground i.e. object of interest in video frames. The other approach i.e. frame differencing is preferred when slow moving objects are required to be detected in the video sequence. This method suffers from the problem that when the object movement is faster then the errors in detection are more. In video surveillance frame size of the video is maintained at higher side for better object detection and also for clear visualization of objects by security personnel. Thus it requires large amount of storage memory. The major hurdle in video surveillance is resolution of video frames and storage of such videos. As we know that when one goes for more resolution then for storing one need more memory. The next hurdle is the time required for processing of such videos for object detection and tracking. It needs very large time to do such operations. To solve these two problems, in this paper, the variance based object detection is proposed and is compared with existing well known mean shift method and for video compression, Discrete wavelet transform (DWT) along with embedded zero- tree wavelet (EZW) is used. The outputs of the DWT are used for both compression and object detection. Further some morphological operations are used for filtering out any noise available in the video frames. Variance method can be used iteratively to get more accuracy.