Background Modeling in Video Processing: A Comprehensive Review

Background Modeling in Video Processing: A Comprehensive Review

Background modeling plays a crucial role in various video processing applications, including surveillance, object detection, and video compression. This paper reviews several significant research efforts in this domain, highlighting the diverse applications and advancements of background modeling techniques.

Lyu et al. [2] proposed a visual early leakage detection system for industrial surveillance environments. Their system utilizes an established background model to effectively extract dynamic potential leakage foreground, demonstrating the efficacy of background modeling in detecting anomalies.

In the realm of surveillance video coding, Li et al. [5] introduced a novel approach to expedite the search process by leveraging a background model. Their method involves an initial background modeling step, followed by the implementation of 'CU Classification' based on the established background. This approach showcases the potential of background modeling in enhancing the efficiency of video coding algorithms.

Wang et al. [6] further emphasized the significance of surveillance video coding in improving compression efficiency for intelligent video surveillance systems. They proposed a background modeling and referencing scheme specifically designed for moving cameras-captured surveillance video coding in high-efficiency video coding (HEVC). This scheme comprises a low-complexity motion background modeling algorithm and utilizes motion background coding tree units (MBCTUs) to update the coding tree unit in the background reference picture's global compensation location. Experimental results demonstrated remarkable bit savings of up to 26.6% and an average of 6.7% with comparable subjective quality and negligible encoding complexity compared to HM12.0.

Advancements in deep learning have also influenced background modeling techniques. Tezcan et al. [7] presented a novel supervised background subtraction algorithm for unseen videos (BSUV-Net) based on a fully-convolutional neural network. This algorithm utilizes the current frame and two background frames captured at different time scales along with their semantic segmentation maps as input. This work highlights the successful integration of deep learning with traditional background modeling principles.

The authors argued that the advancements in deep learning within computer vision did not overshadow background subtraction (BGS) algorithms, which fundamentally rely on background modeling. The wide range of applications for background modeling technology in video processing has motivated numerous researchers to dedicate their efforts to this field, resulting in continuous innovation and development.