Deep neural network concepts for background subtraction:A systematic review and comparative evaluation.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Conventional neural networks, like Self-Organizing Background Subtraction (SOBS), were effective for background subtraction with static cameras.
• Deep learning, especially convolutional neural networks (CNNs), has become prevalent in background subtraction, outperforming traditional methods.

Purpose of the Study:

• To provide the first comprehensive review of deep neural network concepts applied to background subtraction.
• To analyze the reasons behind the success of deep learning in this field.
• To offer directions for future research.

Main Methods:

• Surveyed background initialization and subtraction methods utilizing deep neural networks.
• Analyzed deep learned features relevant to background subtraction.
• Discussed the suitability of deep neural networks for background subtraction tasks.

Main Results:

• Deep neural network-based methods show significant performance improvements over conventional unsupervised approaches.
• Recent top-performing methods on the CDnet 2014 dataset are based on deep neural networks.
• Continuous performance gains have been observed since 2016.

Conclusions:

• Deep neural networks are highly effective for background subtraction, offering substantial advancements.
• Further research is warranted to explore and optimize deep learning applications in this domain.
• This review serves as a guide for both novices and experts in the field.