Multi-scale based enhanced ViBe for background modeling
Zhu Zhenfen · Journal of Beijing Jiaotong University · 2015
A multi-scale based enhanced ViBe named mViBe is proposed.Applying the multi-scale strategy to select the candidate pixels for building the background model,the pixel's spatial information is fully exploited together with the color information to beneficially boost the discriminative ability of the background model.In addition,to improve the background model's robustness to complex scene's changes such as multi-modal background,camera jitter,and illumination change,a hierarchical model updating method based on multiple prior probabilities is proposed.Compared with ViBe and other representative algorithms on some testing videos with complex scenes,the proposed mViBe algorithm shows competitive detection performance without loss of the efficiency.