Alongside autonomous submissions, foreground extraction is considered to be the foundation for various video content analysis technologies, like moving object tracking, video surveillance and video summarization. This paper proposes an efficient foreground extraction methodology based on conditional Generative Adversarial Network. The proposed generator, which is made up of two networks working in series- Foreground Extractor and Segmentation Network, maps the video frames to corresponding foreground masks. The discriminator aids the learning of generator by learning to differentiate between seemingly real and fake foreground maps. The method used a multi-scale approach in order to capture robust features across multiple scales of input using the Feature Extractor Network, which are then used by the successive Segmentation Network to produce the final foreground map. In addition, a multi-frame approach is also used to facilitate capturing of appropriate temporal features. The performance of the proposed model is evaluated on CDnet 2014 Dataset and outperforms existing methods.