基于颜色和局部二值相似模式的背景减除

Background Subtraction Based on Color and Local Binary Similarity Pattern

针对视觉背景提取(Visual Background Extractor,ViBe)算法对光照变化和动态背景适应性差、鬼影消除时间长的缺点,提出了一种改进的ViBe算法。该算法使用颜色特征与局部二值相似模式(Local Binary Similarity Pattern,LBSP)特征进行背景建模,增加了对光照变化的鲁棒性。在模型更新阶段,引入二次空间传播机制,以加快消除鬼影的速度。根据当前像素与空间邻域像素的标准差获得自适应判决阈值,以较快的时间响应速度抑制动态背景的干扰。在Change Detection dataset数据集上的实验结果表明,改进后的算法能较快地抑制鬼影,同时能保证慢速和静止目标不会很快地融入背景,对复杂动态场景和光照变化有较好的适应性,其F-measure指标较ViBe算法提升了19.29%。

As the ViBe（Visual Background Extractor） algorithm is not adaptable enough for illumination variation and dynamic background,an improved ViBe algorithm was proposed.The new algorithm builds the background model with color feature and local binary similarity pattern （LBSP） against illumination variation.A twice spatial diffusion process is introduced to speed ghost-eliminating in the model-update phase.A self-adaptive threshold is obtained via the standard deviation of the current pixel and its neighborhoods in order to inhibit disturbances from dynamic background with a more quick response.Experimental results on the Change Detection dataset show that the new algorithm can rapidly suppress ghosts while keeping a slow inclusion of real static foreground objects and can adapt to complex dynamic background and illumination variation.Compared with the ViBe algorithm,its F-measure is improved by 19.29％.