一种基于反馈信息的视觉图像背景建模方法

A GMM background modeling algorithm based on feedback information

高斯混合模型已经广泛应用于视觉图像的运动目标提取.但传统高斯混合模型存在静止前景融入背景的问题.为了解决这个问题,提出了一种特定场景下基于反馈信息的背景模型更新改进算法.首先采用基于形状特征的目标分类器将前景目标识别为行人和车辆,然后通过多目标跟踪判断目标是否静止,进而将前景目标识别为静止行人,运动行人,车辆三种模式,最后将跟踪与分类的结果与高斯模型的更新相结合,根据分类后反馈的信息对不同的分类区域采取不同的学习率更新.实验结果表明,该方法能够有效地解决特定场景中前景融入背景的问题.

Gaussian Mixture Model has been widely used in video object extraction. However, the problem in traditional GMM is that the still foreground pixel are often blended in back- ground pixel. To solve this problem, this paper proposed a novel rate control scheme based on feedback information. First, the proposed method divided the objects into pedestrians and cars. Second, a multiple object tracking algorithm is proposed to determine whether the target was sta- tionary, then the detected objects can be classified to still pedestrians, moving pedestrians and cars. At last, different regions are adopted to different learning rate depending on the feedback of the tracking and classification results. Experiments show the improved algorithm can solve the problem of still foreground pixel blended in background pixel.