Foreground segmentation is an important part of computer vision and has many useful applications, such as tracking objects and detecting anomalies. However, deep learning models often struggle with the difficult task of preserving small details when extracting features, which is crucial for accurately segmenting edges. To overcome this challenge, we have developed a new segmentation network with a cascading architecture. This innovative approach gradually incorporates finer details into higher-level features, continuously improving the segmentation results. We have thoroughly evaluated our method using the challenging CDnet2014 dataset, and it has shown exceptional performance with an impressive F-measure of 0.9868. This research demonstrates the potential of cascading networks to greatly improve foreground segmentation in computer vision applications, and it is completely original work.