Abstract: Aim at the disadvantages of traditional mixture Gaussian model in moving object detection, an improved moving object detection method based on mixture Gaussian model is proposed. It solves the problem affected by the illumination mutations and the traditional frame difference is easily affected by double shadow which combines the improved mixture Gaussian model and four-frame subtraction. The improved mixture Gaussian model adjusts the numbers of the Gaussian distribution adaptively and improves the accuracy of the background description. This paper discusses the motion state of the object and different learning rates are set to improve the effect of slow-moving object detection. Experimental results show that the combined algorithm can detect moving object accurately and has better adaptability in complex scenes.

Key words: mixture Gaussian model, moving object detection, four-frame subtraction, learning rate

摘要： 针对传统混合高斯模型检测运动目标中存在的不足，提出了一种改进的基于混合高斯模型的运动目标检测算法。将改进的混合高斯模型与四帧差分相结合，有效地解决了突变光照的影响并消除了传统帧差法检测目标时容易出现的双影现象，改进的混合高斯模型自适应地调整了高斯模型的分布数量，提高了背景的描述精度。分情况讨论了物体的运动状态并分别设置不同的学习率，改善了对运动缓慢目标的检测效果。实验结果表明结合后的算法能对运动目标进行准确检测，对复杂场景有较好的适应性。

关键词: 混合高斯模型, 运动目标检测, 四帧差分, 学习率