摘要
讨论了背景模型的更新参数与模型精度的关系。通过精确的梯度背景模型值间接估计当前帧中背景像素理论上的期望梯度值。以高斯模型为基础,将当前帧背景像素的实际梯度值与其理论上的期望值进行比较,计算偏差概率,以此为基础,形成不依赖于局部纹理的梯度特征的相似性度量方法。再用梯度特征的相似度量化地调整差分图像在各像素点处的二值化阈值,实现像素值信息与梯度信息的融合使用。实验表明,本方法对前景分割有一定的改善效果。
The relation between accuracy and the study speed of the background model was discussed. The theoretical gradient expectation was estimated with the accurate gradient background model. Based on Gaussian model, the proba- bility of the deviation between the actual gradient and its expectation was given, leading to a similarity measurement of the gradient feature using no texture message. The similarity was then used to adjust the threshold for binarization of the difference image, which means the fusing use of grey level message and the gradient message. Experiments show that the proposed method does have some improvement on foreground segmentation.
出处
《计算机科学》
CSCD
北大核心
2015年第8期300-304,共5页
Computer Science
关键词
背景减除法
噪声
高斯模型
梯度特征
相似度
预计算
Background subtraction, Noise, Gaussian model, Gradient feature, Similarity, Pre-calculation
