相机抖动场景下的运动前景检测算法

A moving foreground detection algorithm under unstable camera

前景检测是视频监控中信息提取的关键,而相机抖动造成背景边缘的像素极易误检为前景像素,降低前景检测的精确度.为此,提出相机抖动场景下一种基于运动信息的前景检测算法:分析二值图像中候选前景点的运动信息,构建非参数的背景运动信息分布模型;计算候选前景的运动信息与背景模型的概率似然性,由自适应的阈值控制来确定真实前景,该自适应阈值由Mean-shift及信息熵算法共同确定,可以克服单个的全局阈值对场景变化适应能力差问题;针对检测到的前景点和背景点的运动信息,采用首进首出的策略更新背景运动信息分布模型,提高模型对场景实时变化的适应性.实验结果表明,该算法具有良好的鲁棒性,能有效地检测相机抖动场景下的运动前景.

Foreground detection is a fundamental step of extracting information in many visual surveillance applications,but background edge pixels are mistakenly identified as foreground pixels,which reduces the foreground detection accuracy.So a foreground detection algorithm based on motion information is proposed in this paper.Firstly,motion information of the candidate foreground pixel in the binary image is analyzed and a nonparametric model of background motion information distribution is constructed.Then,the likelihood probability between motion information of the candidate foreground pixel and the model is calculated.And the real foreground is determined by an adaptive threshold,which is estimated utilizing Mean-shift and information entropy.By using the adaptive threshold,the approach can overcome defects of using only one global threshold.Finally,according to the detected foreground and background motion information,the background model is updated by using a first-in first-out manner.The experimental results demonstrate that the proposed algorithm is suitable and effective for foreground detection in camera jitter scenes.