摘要
基于背景建模的运动目标分割是智能视频监控的重要任务,模型的质量直接影响到检测、跟踪、识别等运动分析的准确性.当前的建模方法多是单层的,忽略了像素特征在时域和空域上的联系,模型描述不够准确,对于背景扰动、全局光照变化及复杂的室内外场景等多种情况鲁棒性不强,导致了分割中出现空洞和噪声点.针对这些问题提出了一种双层建模的方法,在第一层提取时域上的像素亮度特征采用码本建模,第二层提取邻域纹理特征采用基于中心对称的局部二值模式建模.实验证明该方法在用于运动分割时,比常用方法具有更好的准确性和鲁棒性.
Moving object detection using a background model is one of the most important targets for video surveillance.The quality of the model affects the accuracy of object detection,tracking and recognition.Most of the traditional methods are single layered,with no consideration of the change in brightness both spatially and temporally.The model is therefore not accurate enough to handle difficulties such as waving trees,light switching,complex background and there will be holes and noise when segmentation is done.A two-layer method of spatio-temporal background modeling is described.A modeling algorithm based on codebook（CB） was presented in the pixel-level layer,while a texture-based algorithm using center-symmetric local binary pattern（CSLBP） in the region-level layer.Experimental results on video sequences demonstrate that the method can process movement segmentation effectively.
基金
安徽省科技攻关项目(09010306042)资助
