基于帧间相似性约束鲁棒主成分分析模型的运动目标检测  

MOVING TARGET DETECTION BASED ON RPCA WITH INTERFRAME SIMILARITY CONSTRAINT

作　　者：杨国亮[1] 鲁海荣[1] 丰义琴 黄经纬[1] 

机构地区：[1]江西理工大学电气工程与自动化学院,江西赣州341000

出　　处：《计算机应用与软件》2016年第1期142-146,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(51365017;61305019);江西省科技厅青年科学基金项目(2013bab211032)

摘　　要：针对鲁棒主成分分析模型RPCA(robust principle component analysis)未能有效地利用相邻两帧具有相似性这一特性,提出基于帧间相似性约束鲁棒主成分分析模型的运动目标检测算法。考虑到时间序列数据中相邻数据之间的相似性特性,在原始的RPCA模型基础上,引入帧间相似性约束条件,通过求解新的RPCA模型可以得到平滑的低秩数据矩阵和稀疏误差矩阵,有效保留了原有序列数据中的相似性结构。将该模型用于运动目标检测,观测图像序列分解成低秩背景矩阵和稀疏运动目标矩阵,对分解出的运动目标进行二值化,并对检测出的运动目标图像进行定性分析和采用Similarity与F-measure评判标准进行定量分析。通过实验结果分析,该算法能够有效地对运动目标进行检测,提高运动目标的检测率。Robust principal component analysis （RPCA） model does not effectively utilise the characteristic that the adjacent two frames has the similarity, for this issue, this paper proposes a moving target detection algorithm which is based on restraining RPCA with interframe similarity. Considering the similarity feature of adjacent data in time series data, based on primitive RPCA model we introduce the constraint condition of interframe similarity, by calculating new RPCA model it is able to obtain smooth low-rank data matrix and sparse error matrix, thus effectively preserves the similarity structure in original series data. We applied this model in moving target detection, observed the decomposition of image series into low-rank background matrix and sparse moving target matrix, and conducted the binarisation on the decomposed moving target, as well as made the qualitative analysis on the detected moving target image and adopted Similarity and F-measure judgement criteria for quantitative analysis. Analysis on experimental results showed that the proposed algorithm could effectively detect moving target and improved the detection rate of moving target.

关 键 词：鲁棒主成分分析 序列数据 帧间相似性约束 运动目标检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]