长期视频监控系统的多分布模型背景差方法

MULTI-DISTRIBUTION MODEL FOR BACKGROUND SUBTRACTION IN LONG-TERM VIDEO SURVEILLANCE SYSTEM

提出在长期视频监控系统中采用背景差进行运动目标提取时算法所要满足的基本要求 ,并提出了一种能够满足这些要求的背景差方法 .该方法用色度、亮度空间的多个分布模型来建立背景模型 ,描述彩色视频图像的背景像素点及其统计特性 ,在对背景模型更新时将均值、方差的更新速率和多个模型的更替速率分开 .对像素值属于多个分布模型的情况 ,用最小相似距离确定要更新的模型 .该方法利用提取的前景像素点信息反馈以检测光强的突变 ,利用亮度信息消除运动目标的阴影 .

The basic performance requirement of exploiting background subtraction approach in long-term color video surveillance system was described. To meet the requirement, an algorithm was also proposed. In this approach, by using the multi-distribution models in lightness and chromaticity spaces, the background model was built. The multi-distribution models were then updated using independent mean and covariance updating rate and model replacing rate. When a pixel could be represented by more than one distribution model, the model which has the minimum similarity distance was updated. The approach also involves the detection of the sudden changes in illumination by the feedback of the information of foreground pixels and the suppression of shadow influenced by lightness information. Experiments show that the method satisfies the demands of long-term video surveillance.