【摘要】 依据矩阵的低秩表示可以恢复矩阵主要信息并抑制噪声干扰的原理,提出一种适用于动背景条件下的运动检测方法。在背景建模阶段,采用傅里叶模式将实际场景的视频序列矩阵近似分解成一个背景部分的低秩分量和前景部分的稀疏分量,依据低幅值的傅里叶模式重建背景模型,有效克服动背景引入的背景模型噪声;在前景检测阶段,依据两幅低幅值傅里叶模式的重建图像自适应计算前景分割阈值,有效分离前景和背景。在Changedetection.net的7个数据子集上进行仿真实验,其结果表明,采用所提方法进行运动检测的正确分类百分比指标和处理帧率指标高于常用的VIBE和GSOM方法。更多还原

【Abstract】 According to the principle that low-rank representation of a matrix can recover the main information of the matrix and suppress noise interference,a motion detection method for dynamic background was proposed.In the phase of background modeling,the video sequence from actual scene was approximately divided into two parts using Fourier modes.One was the background with low-rank component,and the other was the foreground with sparse component,and background model according to the low amplitude of the Fourier modes was reconstructed,to effectively overcome the noise of background model from dynamic background.In the phase of foreground detection,the segmentation threshold was adaptively computed according to two images reconstructed using two Fourier modes with low amplitude,to separate the foreground and background effectively.Results of simulation experiments on seven sub-datasets of Changedetection.net show that two measures including percentage of correct classification and frames per second for motion detection of the proposed method are higher than commonly used methods such as VIBE and GSOM.更多还原