Background Modeling in Deep Learning Frameworks: A Response to Reviewer Feedback

We would like to express our gratitude for the valuable feedback provided on our paper publication. We acknowledge the suggestion put forth regarding the use of a deep learning framework for background modeling. While it is widely recognized that deep learning algorithms have demonstrated remarkable potential in various image processing tasks, we have observed that traditional techniques are not in conflict with deep learning frameworks.

For instance, in the publication 'BSUV-Net: A Full Convolutional Neural Network for Background Subtraction of Unseen Videos' by Tezcan M. O. et al. (Disclaimer: Neither I nor my colleagues are affiliated with this work), the authors propose the BSUV-Net Algorithm for unseen videos, which is based on a full convolutional neural network. In BSUV-Net, the authors utilize two reference frames to characterize the background. One of the frames is an Empty 'background frame,' which depicts the background without any people or objects of interest. The other reference frame characterizes the recent background, for instance, by calculating the average of the 100 frames preceding the frame being processed.

This signifies that the background generated by traditional methods can also be incorporated as a component of deep learning frameworks, and the quality of the background generated significantly influences the results of deep learning. Therefore, it is meaningful to construct the background based on traditional methods in this article.