Abstract: An adaptive approach to detect moving objects with a static camera was proposed in this paper. Based on the assumption that the probability of the same pixel covered by the same intensity of the car is the least, the initial background was established and the number of Gaussians for each pixel on-line was chosen. In order to update the Gaussians parameters, the learning rate should be updated online according to the relationship between the pixel and its adjacent pixels. Lastly,the moving objects were detected by background subtraction. Compared to the moving objects detection approach based on conventional mixture Gaussian model, this approach has preferable adaptive performance, and can deal with changes in scenes rapidly.

Key words: mixture Gaussian model, background updating, background subtraction, objects detection, noise removing

摘要： 提出了一种静止摄像机条件下自适应的运动目标检测方法。该方法基于同一像素点被同一灰度车辆覆盖几率小的假设构建初始背景，为每个像素点在线选择高斯分布个数。根据像素点与其邻域像素间存在联系的思想，在线更新学习率。最后用背景差分法检测出运动目标。实验结果表明，同基于传统混合高斯模型的运动目标检测方法相比，该方法有较好的自适应性，能快速适应场景的变化。

关键词: 混合高斯模型, 背景更新, 背景差分, 目标检测, 噪声去除