基于平滑帧差法的混合高斯模型的研究与应用

Research and application of Gaussian mixture model based on smooth frame difference

鉴于传统混合高斯模型在光照突变、噪声干扰时鲁棒性不高,易造成检测错误等问题,提出一种改进的监控视频前景运动目标提取算法.该算法将帧差法平滑化,再与混合高斯模型相结合,分别采用正序和倒序对样本灰度矩阵处理,增强识别准确度.通过基于灰度图的协方差矩阵导出仿射不变量,根据已有的仿射不变量对灰度图进行识别、分析,得到在晃动视频下的前景运动目标的较准确提取.不同场景的视频检测结果表明,改进算法有效克服了监控摄像头晃动或偏移、光照突变、噪声干扰、"空洞"及"双影"现象,与同类算法相比,具有更高的准确度和鲁棒性.

In view of the problem that the robustness is not high during light mutation and noise interference of the traditional Gaussian mixture model,which easily causes detection errors and other problems,this paper proposes an improved algorithm for foreground moving object detection of video surveillance.The algorithm smoothes the frame difference method.And then combined with the Gaussian mixture model,the algorithm respectively uses the positive sequence and the reverse sequence to the sample gray matrix to enhance the recognition accuracy.The affine invariant is derived from the covariance matrix based on grayscale and the grayscale image is identified and analyzed based on the existing affine invariant,so as to obtain a more accurate extraction of the foreground moving object under shaking video.The video detection results of different scenes show that the improved algorithm effectively overcomes the phenomenon of camera shake or shift,light mutation,noise interference,cavity and double shadow.Compared with other similar algorithms,it has higher accuracy and robustness.