Background Subtraction Model based on Adaptable MOG

Mixture of Gaussian (MOG) approach is a powerful estimation and prediction background subtraction model. Nevertheless, although it has been improved by using several algorithms such as Expectation Maximization (EM); it is still susceptible to sudden changes in light conditions effects. In this paper, we analyze the MOG approach in order to explore its strengths and weaknesses in order to create a new robust algorithm. Our proposal consists on a new algorithm based on a dynamic selection of convergence ratio, which use the expected proportion between movement and fixed zones of scene. This proportion is used as an extra criterion to detect the maximum direction of Entropy in EM algorithm. The algorithm suits best convergence ration due to global changes in scene. Finally, in an experimental model, our approach is tested in outdoors and indoors scenarios, where luminance conditions has changed. Results show the adaptability of our approach to several dynamic scenarios.