基于块模型的混合高斯模型运动目标检测方法

A moving object detection method based on block model gaussian mixture model

文中针对传统混合高斯模型(GMM)运动目标检测方法计算量大、时间复杂度高的缺点,提出一种利用块模型的混合高斯模型运动目标检测方法。该改进算法利用分块处理技术为每个块建立一个模型,同时利用概率更新策略对块模型进行更新,充分利用图像像素间的空域信息,大量减少算法的计算量和存储空间,提高了算法的运行效率。应用这种改进算法,对分辨率不低于CIF(352×288)的监控视频进行检测,结果表明:当块大小值设置为3×3时,检测效果与传统混合高斯模型的检测效果基本一致,而改进算法的平均耗时减少了46.16%,存储空间减少不低于54.15%。

The traditional Gaussian mixture model （ GMM） moving object detection method is high （computation and high time complexity. In this paper, a novel moving object detection method based on block model Gaussian mixture model is proposed. It utilizes block processing technology to build a model for each bloci, and updates bloci models by probability update strategy. Therefore, the improved way can make full use of spatial iniormation of the image pixels, and significantly reduces computational complexity and storage space. By applying the proposed algorithm to videos with no less than the CIF （352× 288） resolutions, the experiment shows that when the block size value is set to 3 × 3 pixels, detection results are almost same compared with the traditional Gaussian mixture model, while 46. 5 7 % processing time and more than 54. 15% storage space are saved.