摘要
针对原始码本模型算法参数多且难以设定、背景模型更新缓慢、背景容易被误判为前景目标等问题,本文提出一种改进的码本模型背景建模算法。该算法以像素点的均值和方差作为码本背景中码字的参数,用高斯正态分布判断该像素值是否匹配某个码字,采用以码字出现频率为参数的动态更新率来更新高斯正态分布中的均值及方差;采用两层的背景建模技术添加与删除背景码本中的码字;并增加频率阈值来判断某个像素值对应的码字是否添加入/删除出背景码本模型或缓冲区码本模型。实践证明,该算法能在复杂的场景下准确的分割出运动目标,且快速实用,能应用于实时视频图像处理中。
As an effective background subtraction algorithm,the basic codebook model algorithm has many disadvantages.Such as too many parameters background was mismatched as foreground,the slow speed of the background model updating,affected by noises.An improved codebook model algorithm is proposed.The mean and variance cubic（MVC） of the pixel value in RGB color space was applied to judge whether the input matches one of the codewords or not; the mean and variance of codeword is updated by dynamic learning rate which is correlated to the frequency of the codeword has occurred; layered modeling technique was proposed to update codebook adaptively; meanwhile the frequency of the codeword has occurred was also used as one of the criteria to judge whether absorbing a codeword into the codebook model.Experiment results show that the algorithm proposed can segment moving objects under complicated scenes which could include dynamic background and illumination change,can absorb variations into the background model quickly,can be used to real-time foreground-background segmentation
出处
《科学技术与工程》
北大核心
2017年第32期148-153,共6页
Science Technology and Engineering
基金
国家自然科学基金(61503141)资助
