Moving object detection is a core task in computer vision. However, existing deep learning-based moving object detection methods require a large number of labeled frames to achieve good generalization and performance. In moving object detection tasks, there is no such a large-scale labeled dataset because the labeling process requires a lot of time and effort. In this paper, we compiled a large-scale dataset by 1) combining existing moving object detection datasets. 2) using an in-painting deep learning model to transform datasets from video object segmentation task to moving object detection tasks. 3) generating synthetic datasets by combining random backgrounds with random foreground objects. Additionally, we propose a novel deep-learning model that performs background subtraction on the object level. This model is trained on the compiled dataset and shows superior performance in the moving object detection task. The model is evaluated using the CDNet dataset and results are compared with current state-of-the-art models. The results show that our model outperforms the best-reported state-of-the-art model by \(1.6\%\) .