基于箱式约束RPCA的运动目标检测

为克服运动目标检测中光照变化、阴影干扰等的影响,提出了一种具有箱式约束的鲁棒主成分分析方法,用于带阴影的视频运动目标检测。该方法建模时首先将输入的视频数据分解为低秩背景、稀疏前景与阴影3个部分;接着在传统鲁棒主成分分析模型的基础上对阴影变量施加箱式约束,利用Powell-HestenesRockafellar增广拉格朗日乘子法将上述约束转化为目标函数的惩罚函数项,推导了3个子问题的闭合解,并用交替方向法对模型进行求解;最后在公开数据集上对该方法进行了测试。实验结果表明,该方法能够在检测运动目标的同时去除阴影,场景适应性较好。

Moving object detection via box constrained RPCA

To address the issue of illumination change and shadow challenges in moving object detection, a new robust principal component analysis (RPCA) method with box constraint (BC-RPCA) was proposed to separate the foreground with shadow around in videos. First of all, the method modelled the input image sequence as three parts which are low rank background, sparse foreground and moving shadow. Then box constraints were casted on the basic RPCA model and Powell-Hestenes-Rockafellar (PHR) Augmented Lagrangian method was employed to form the objective function. Closed-form solutions to three sub-problems were also conducted and alternating direction method of multipliers (ADMM) used to solve the problem effectively. Finally, the experiments on several scenes demonstrate the proposed method works well on shadow and illumination change challenge.