基于交替方向低秩模型的运动目标检测算法

A MOVING OBJECT DETECTION ALGORITHM BASED ON ALTERNATING DIRECTION AND LOW-RANK MODEL

不同于传统运动目标检测算法,引入背景低秩和前景稀疏性,提出基于交替方向低秩模型的运动目标检测算法。首先在鲁棒主成分分析法建模的基础上添加背景噪声模型,在低秩背景模型中引入全变差范数并结合核范数进行约束。考虑视频矩阵前景图像的稀疏性,接着利用马尔可夫随机场和图建立前景模型。然后采用交替方向法实现函数的优化求解。最后对算法结构进行改进,实现视频运动目标的在线检测。通过对两种数据集进行实验结果分析,与其他算法对比,该算法在满足在线的基础上具有很好的检测效果,特别是在动态背景及复杂前景上具有很强的鲁棒性。

Different from traditional moving object detection algorithm, we propose a moving object detection algorithm by introducing background low-rank and foreground sparsity, which is based on alternating direction and low-rank model. First, it adds a background noise model based on modelling the robust principal component analysis （ RPCA）, introduces total variation ttorm to low-rank background model and combines nuclear norm to constrain. Secondly, it builds the foreground model by using Markov random field and graph with the consideration of the sparsity of foreground image in video matrix. Then, it uses alternating direction method to realise the optimised solution of the function. At last, we improve algorithm＇ s structure to implement the online detection of moving video objects. Analyses on the results of experiments on two different datasets tell that based on online satisfaction this algorithm has better detection effect compared with other algorithms, in particular, it has strong robustness in dynamic background and complex foreground.