混合高斯模型与三帧差分法相结合的建模新算法

New algorithm based on Gaussian mixture model and three frame difference method

针对三帧差分法所存在的不足,提出一种改进的基于混合高斯模型与三帧差分相结合的建模方法。为每一个背景像素建立多维混合高斯模型,融入三帧差分法实时判定背景区域和运动区域,使之去除三帧差分带来的空洞现象,并且加入可跟随目标移动的外接矩形框,在其内生成高斯模型,从而减少因高斯模型的介入导致计算量过大的问题,节省运算时间,并且达到理想的除噪效果以及排除外界不必要的干扰等。通过实验进行验证分析,实验结果表明:该方法相比三帧差分法具有更好的除噪效果和减少更多的计算量,适用于实时的单目标检测。

An improved model based on Gaussian mixture model is proposed which is combined with three frame difference method,aiming at the shortcomings of the three frame difference method. We establish multidimensional Gaussian mixture model for each background pixels to remove the cavitations and add three frame difference method to Real-time determine the background area and sports area. An external rectangular box is added which can follow the target ＇s movement and in which Gaussian mixture model is generated. This method can reduce the large calculating quantity and save operation time for the intervention of Gaussian mixture model,get ideal effect of noise cancellation and eliminate the unnecessary interference. Through the experimental validation and analysis,it turns out that compared with three frame difference method,this method has better effect on noise cancellation and can decrease more computing. It is suitable for the real-time target detection.