基于时空的混合高斯背景建模的运动目标检测

本文针对摄像机固定下的复杂背景环境，提出一种基于时空的自适应混合高斯背景建模方法，克服经典混合高斯模型（Gaussian Mixture Model，GMM）中只考虑单个像素的独立性而忽略相邻像素间的空间域相关性。首先采用混合高斯模型对每个像素在时间域上进行学习，然后利用相邻像素的自信息对背景及前景目标进行二次聚类，以修正错误的判断。实验结果表明，与经典混合高斯背景算法相比，本文提出的方法目标检测结果更加完整，具有更强的鲁棒性和很好的应用前景。

Moving Object Detection Based on Temporal-spatial Mixture Gaussian Background Model

This paper proposes a temporal-spatial mixture gaussian background model which overcome the standard GMM where each pix is only considered independently but ignoring the spatial domain correlation between neighboring pixels. Based on the temporal distribution model learned by gaussian mixture model, self-information of neighboring pixels are clustering for background and foreground object to correct an error of judgment. Experimental results that detected objects with the proposed is more completed, and has stronger robustness application prospects.