Parallel Implementation of Gaussian Mixture Model Background Subtraction on Jetson Nano

Abstract

Background subtraction (BS) is an essential component in numerous video processing tasks, including object tracking and recognition. The Gaussian Mixture Model (GMM) is one of the most popular techniques for BS, but it's highly computationally demanding. This paper addresses this challenge by proposing a parallel implementation of GMM on Jetson Nano's quad-core ARM CPU. We leverage Open Multi-Processing (OpenMP) for efficient parallelization and evaluate the performance using various scheduling approaches, including static, dynamic, guided scheduling, and orphan directive. Furthermore, we investigate the impact of chunk size on parallel efficiency within each scheduling approach. Our evaluation demonstrates significant improvements in processing speed, particularly for high-resolution frames. The parallel efficiency reaches 92.5% for full high-definition resolution frames and 85% for low-resolution frames, respectively.