Abstract:

Aiming at the background modeling in complex scenes,this paper proposes a novel adaptive fuzzy method based on function estimation.In this method,the Takagi- Sugeno- Kang (TSK) fuzzy system is taken as the estimator,and the parameters of the premise part and the consequent part of the fuzzy system are optimized by combining the particle swarm optimization (PSO) with the recursive least squares estimator (RLSE).In order to effectively esti-mate the background,the foreground samples are interpreted as outliers relative to the background samples,and an outlier separator method is devised.After the outliers are removed,the obtained results are used to train the fuzzy estimator.Finally,through the experiments of different video sequences,it is found that the proposed method is accurate and effective in such enviromnets as dynamic background,illumination changes and camera vibration.

Key words: background modeling, TSK fuzzy system, particle swarm optimization, recursive least squares estimator