An optimized parallel implementation of the codebook algorithm enables real-time background subtraction at speeds of up to 370 frames per second for high-definition video and 172 frames per second for Full-HD resolution. Detailed in Multimedia Tools and Applications, the experimental results demonstrate that the new CPU and GPU implementations significantly outperform existing state-of-the-art codebook-based benchmarks.
Background subtraction represents a critical component across diverse computer vision systems, serving as an essential processing stage in video surveillance and intelligent transportation applications. Within this domain, the codebook algorithm has seen widespread adoption because of its inherent robustness. Although researchers have previously introduced several algorithmic modifications to enhance detection accuracy, these methods typically impose substantial computational overhead that renders them impractical for operational real-time use.
This computational performance barrier primarily originates from the algorithm’s structural complexity and the variable number of codewords generated per pixel during execution. Because the quantity of codewords continually changes across an image, standard implementations must perform frequent and costly dynamic memory allocations.
To overcome these computational limitations, the authors designed an optimized codebook algorithm formulated to resolve the underlying memory management challenges. The proposed approach significantly reduces the overall memory footprint of the background model while strictly preserving the algorithm’s baseline detection accuracy, thereby preventing the processing slowdowns that afflicted earlier accuracy-enhancing variants.
The research team coupled this streamlined algorithm with parallelization techniques specifically tailored for two separate hardware environments: a conventional laptop equipped with an NVIDIA GeForce RTX 3050 graphics processing unit and an NVIDIA Jetson Nano embedded platform. By integrating Open Multi-Processing (OpenMP) with CUDA, the parallelization framework efficiently leverages both available CPU and GPU resources to achieve high-throughput processing across standard and resource-constrained embedded systems.
Subject of Research: Parallel background subtraction algorithms on CPU and GPU platforms
Article Title: Efficient parallel implementation of the Codebook algorithm for background subtraction on heterogeneous CPU/GPU platforms
Article References: Tebbaa, B., Klilou, A., Bariko, S., & Arsalane, A. (2026). Efficient parallel implementation of the Codebook algorithm for background subtraction on heterogeneous CPU/GPU platforms. Multimedia Tools and Applications, 85(12), Article 855. https://doi.org/10.1007/s11042-026-22020-z
Image Credits: AI Generated
DOI: 10.1007/s11042-026-22020-z
Keywords: background subtraction, codebook algorithm, parallel computing, CUDA, OpenMP, graphics processing units, central processing units, heterogeneous computing, computer vision, video surveillance, intelligent transportation systems, NVIDIA Jetson Nano, NVIDIA GeForce RTX 3050, dynamic memory allocation, embedded systems, real-time throughput
News Source: Denise Maddox. (November 16, 2026). Parallel Codebook Algorithm Speeds Background Subtraction to 370 Frames per Second. Scienmag.



