Visual surveillance aims to detect a foreground object using a continuous image acquired from a fixed camera. Recent deep learning methods based on supervised learning show superior performance compared to classical background subtraction algorithms. However, there is still room for improvement in the static foreground, dynamic background, hard shadow, illumination changes, camouflage, etc. In addition, most of the deep learning-based methods operate well in environments similar to training. If the testing environments are different from training ones, their performance degrades. As a result, additional training in those operating environments is required to ensure good performance. Our previous work, which uses spatio-temporal input data consisting of several past images, background images, and the current image, showed promising results in different environments from training. However, it uses a simple U-NET structure. This paper proposes a data augmentation technique suitable for visual surveillance for additional performance improvement using the same network used in our previous work. In deep learning, most data augmentation techniques deal with spatial-level data augmentation techniques used in image classification and object detection. We propose two data augmentation methods of adjusting background model images and past images. The proposed algorithm improves performance in complex areas such as static foreground and ghost objects compared to previous studies. Through quantitative and qualitative evaluation using SBI, LASIESTA, and our dataset, we show superior performance compared to deep learning-based algorithms and background subtraction algorithms. In addition, it has a 30.2% and 27.9% reduction of false detection rate in the LASIESTA and SBI dataset, respectively, compared to our previous study. Author