Background Extraction Method Based on ESO-LFFDL

Background extraction is a commonly used method for moving object detection in video sequences. Most background extraction methods rely on specific models to handle complex scenarios in different situations, in which the problem of model distortion always arises. In this paper, a local full form dynamic linearization (LFFDL) method based on an extended state observer (ESO) is proposed to address the above problem in background extraction. The LFFDL method is developed to build the dynamic relationship between background images and original images as a parametric data model, which can be adaptively updated with a parameter estimation law. Meanwhile, an ESO is designed to estimate the uncertainty in background extraction, and the estimation value is used as a compensation to improve the robustness. The simulation results indicate that the proposed method can achieve the background and foreground extraction in different scenarios, and the F-measures and precision reach over 95%, which is better than other existing methods.