Journals / International Journal of Informatics and Applied Mathematics / 2019 / Cilt: 2 - Sayı: 1
Background subtraction based on a Self-Adjusting MoG
- Pages
- 73–84
- DOI
- —
Abstract
The diversity in background scenes such as, illumination changes, dynamics of the background, camouflage effect, shadow, etc. is a big deal for moving objects detection methods makes it impossible to manage the multimodality of scenes in video surveillance systems. In this paper we present a new method that allows better detection of moving objects. This method combine the robustness of the Artificial Immune Recognition System (AIRS) with respect to the local variations and the power of Gaussian mixtures (MoG) to model changes at the pixel level. The task of the AIRS is to generate several MoG models for each pixel. This models are filtred through two mecanism: the competition for resources and the development of a candidate memory cell. The best model is merged with the exesting MoG according to the Memory cell introduction process. Obtained results on the Wallflower dataset proved the performance of our system compared to other state-of-the-art methods.
Özet
The diversity in background scenes such as, illumination changes, dynamics of the background, camouflage effect, shadow, etc. is a big deal for moving objects detection methods makes it impossible to manage the multimodality of scenes in video surveillance systems.In this paper we present a new method that allows better detection of moving objects. This method combine the robustness of the Artificial Immune Recognition System (AIRS) with respect to the local variations and the power of Gaussian mixtures (GMM) to model changes at the pixel level.The task of the AIRS is to generate several GMM models for each pixel. This models are filtred through two mecanism: the competition for resources and the development of a candidate memory cell. The best model is merged with the exesting GMM according to the Memory cell introduction process.results obtained on the Wallflower dataset proved the performance of our system compared to other state-of-the-art methods.