Efficient parallelization of GMM background subtraction algorithm on a multi-core platform for moving objects detection
Lhoussein Axel Mabrouk, Sylvain Huet, Dominique Houzet, Saïd Belkouch, Abdelkrim Hamzaoui, Yahya Zennayi · 2018
Gaussian Mixture Model background subtraction (GMM) method is nowadays used in many moving object detection applications. This common approach is performed statistically on each single pixel in the captured frames. Thus, it is well suitable for parallel processing. With the great evolution of multi-core platforms, the parallelization of this algorithm is the most efficient way for its real-time acceleration. In this paper, we propose an efficient multi-threading parallelization of GMM on a 16-cores Intel node using the OpenMP framework. This is carried out by removing data dependencies between different threads which slows down the system; balancing their computational load and avoiding some hidden errors when measuring the performance. The use of a suitable compile environment and options showed that high scalability and linear speedup are achieved even when high number of cores is used.