关键词: 目标检测, 核密度估计, 高斯模型, 联级机制

Abstract: Gaussian model and kernel density estimation model are effective ways for background modeling.Calculation of Gaussian model is simple，however，it suffers from low robustness when there are dynamic scenes and/or sudden lighting changes.Kernel density estimation is robustness but it is too complex to calculate in real-time.A cascade detection mechanism is proposed.Most of the stable pixels are segmented by Gaussian model.After that，for a small part of the pixels that the Gaussian model can not accurately describe are segmented by kernel density estimation model.Experiments confirm that the proposed method is effective to deal with dynamic backgrounds and fast in computation.

Key words: target detection, kernel density estimation, Gaussian model, cascade mechanism