A Hierarchical Approach for Robust Background Subtraction Based on Two Views

In this paper, a hierarchical algorithm is proposed based on two views for background subtraction in image processing. The main idea is to explore the usefulness of the background geometric characteristics of interest in background subtraction for performance improvement. In the proposed approach, the input stereo images are decomposed into different levels using DWT (discrete wavelet transformation), and then an adaptive model is built over sub-bands at different levels in a hierarchical fashion. In particular, when the current pixels do not fulfill the adaptive model at the coarsest level, they will be confirmed farther by a depth based model. In addition, the proposed approach can provide the depth information which is critical to 3D tracking, while performing a task of segmenting the foreground objects. Experiments are conducted to illustrate the effectiveness in the background subtraction.