Abstract: A method for finding moving objects by using random strategy is discussed. There are several innovations: firstly, the background model is initialized by the first frame of the video sequence；secondly, a unique decision rule is established to discriminate the moving pixels and the background pixels；thirdly, when updating the background model, not only updates the sample sequence of the current pixel, but also updates the neighborhood’s sample sequence. The random strategy is used in the process of initialization and updating. The spectral, spatial and the temporal features of pixels are exploited. The proposed method, compared with the running average method and the improved Gaussian mixture model method, is proved to be simple but efficient with lower computational cost and higher accuracy.

Key words: sample selecting, decision rule, background updating, random strategy

摘要： 提出一种基于随机策略进行运动目标检测的方法。方法的主要创新点：（1）利用视频序列第1帧完成背景模型的初始化；（2）建立特定的运动点与背景点判定规则；（3）背景模型更新过程中不仅更新当前像素点的样本序列，同时更新其邻域的样本序列。在背景模型初始化与更新过程中，使用随机策略进行样本序列的更新。该方法利用了像素点的光谱、空间和时间特征，从而提高了检测效果。通过与滑动平均算法、改进的混合高斯模型算法进行实验比较，结果证明该方法是一种运算量小，准确率高，简单高效的运动目标检测方法。

关键词: 样本选择, 判定规则, 背景更新, 随机策略