摘要
智能监控系统中的一个关键问题是在图像序列中准确检测前景目标,但在复杂背景中,这仍然比较困难。将前景检测作为标记问题来处理,提出一种基于多尺度的判别模型,用来学习复杂背景。首先,通过基于像素的方法得到静态背景和运动目标;然后利用一组高斯滤波器组作用于不同的图像空间得到一系列的响应,在图像序列中估计这些响应的概率密度作为特征池,运用AdaBoost算法在特征池中挑选弱分类器组成强分类器,通过分类器获得运动目标中每个像素属于真实前景的置信度;最后,结合前景和背景的时空一致性,利用图分割求解马尔可夫随机场,获得准确的前景。实验结果表明提出的方法能很好地适应各种复杂背景。
A key problem in automated surveillance systems is to detect the foreground accurately in image sequence.However,it is difficult in the dynamic and changed illumination scenes.The foreground detection was considered as a labeling problem and a multi-scale discriminative model was proposed to learn the complex background for foreground detection.The static background could be obtained by the pixel-wise method,firstly.Secondly,the pixels in the moving objects could be classified as the dynamic background and foreground associated with the confidence through a boosted classifier.A Gaussian filter bank with different variances was exploited to form the multi-scale images in the different image spaces,then the feature pool could be obtained by kernel density estimation on the image sequence over time.The boosted classifier was trained by the AdaBoost over the feature pool and the labeled positive and negative data.Finally,Markov random field (MRF) model was used to infer the spatial and temporal coherence over the class labeling for foreground/background segmentation accurately.Experiments tested on the various videos show that the proposed method can be work well on the complex background.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2009年第24期7809-7812,7820,共5页
Journal of System Simulation
关键词
智能监控
背景对消
多尺度
核密度估计
马尔可夫随机场
automated surveillance
background subtraction
multi-scale
kernel density estimation
Markov random fields