一种改进型的时空混合高斯背景建模

针对单固定摄像机的环境,提出一种结合三帧差法的时空混合高斯背景建模方法,改善了传统混合高斯背景建模只考虑单个像素在时间上的特性而忽略相邻像素空间上的相关性,从而对非平稳场景较为敏感的缺点。对于状态变化频繁的点,让其取邻域内的稳定点的灰度值,再与三帧差法的结果进行或操作。实验结果表明,与经典混合高斯背景方法相比,本文的方法能够有效改善背景中出现的非平稳波动,检测出的目标也更完整,并且对阴影的影响,也有一定的抑制作用。

Improved spatio-temporal Gaussian mixture model for background modeling

For single fixed camera environment,an improved spatio-temporal Gaussian mixture model for background modeling combined with three-frame-difference method is proposed.It improves the shortcoming of traditional Gaussian mixture background modeling method that only considers a single pixel of independence and ignores the correlation between adjacent pixels,and sensitive to non-stationary scenes.For those pixels which state change frequently,let them take the value of their neighborhood pixel which state is stable,then do the operation with the result of three frame differencing.The experimental results show that compared with the classical mixed Gaussian background method,the proposed method can effectively improve the background non-stationary fluctuation,and the target be detected is more complete,the method also has some immunity to the impact caused by the shadow.