摘要
对含有动、静态背景的稳定图像处理时,对比了主成分追踪鲁棒主成分分析法(RPCA)、贝叶斯鲁棒主成分分析法(Bayesian RPCA)和高斯混合模型的鲁棒主成分分析法(MoG-RPCA),3种方法对静态背景下的前景提取都较为完整.而动态背景下只有Bayesian RPCA和MoG-RPCA提取出了完整的前景目标,但是Bayesian RPCA计算速度很慢,且不能够处理复杂噪声.所以MoG-RPCA模型更具有对复杂噪声的适应性,动、静态背景情况下均提取出精度较高的前景目标,且具有较快的计算速度.当图像不稳定时,采用改进的MoG-RPCA模型对非稳定拍摄的抖动视频进行前景目标提取,并在第197帧抖动图像中清晰地提取出显著前景目标,且运算速度较快.在为了快速找到目标出现的帧时,对高斯混合模型背景差分法进行改进,利用K-means聚类算法快速得到聚类中心点,然后作为高斯混合模型背景更新时的初始化均值参数,从而提高在复杂场景下前景目标的检测精度.对于多角度追踪任务,不同角度、近似同一地点的多个监控视频图像中前景目标的提取,可采用跨摄像头视角跟踪结果融合的方法,然后对目标进行匹配.
On processing the stable video with dynamic and static background,We compare three theories and methods：Robust principle component analysis（RPCA）,Bayesian RPCA,Model of Gaussian-RPCA,which extract the complete foregrounds under static background.However,just B-PRCA and MoG-PRCA work better under dynamic background,although the B-PRCA can not deal with the complex noise and run slowly.Improved MoG-PRCA could extract the foregrounds fastly at 197 th frame from the unstable video.Gaussian mixture model background difference method（GMMBDM）is improved for finding the significant goal from video,that the method utilize K-means algorithm to get the cluster center point quickly.The GMM-BDM uses the cluster center point as its initialize the mean parameter to flush the background frame,and increasing the accuracy of complex target detection.For multi-angle tracking task,the extraction of foreground objects in multiple surveillance video images from different perspectives and at the same location can be achieved by using the method of tracking the results across camera angles and then matching the targets.
作者
刘钱源
杨欢欢
刘培鑫
张承进
LIU Qianyuan, YANG Huanhuan, LIU Peixin, ZHANG Chengjin(School of Mechanical,Electrical ＆ Information Engineering,Shandong University（Weihai）, Weihai, Shandong 264209, China)
出处
《数学建模及其应用》
2018年第1期63-71,共9页
Mathematical Modeling and Its Applications
