时空RPCA在复杂场景下的运动目标检测

在复杂动态背景下,鲁棒主成分分析模型(RPCA)容易将背景中动态背景误判为前景运动目标,导致运动目标检测精度不高。为解决该问题,提出一种基于非凸加权核范数的时空低秩RPCA算法。使用非凸加权核范数替代传统的核范数进行低秩约束,在观测矩阵上通过拉普拉斯特征映射得到时空图拉普拉斯矩阵,将得到的时空图拉普拉斯矩阵嵌入低秩背景矩阵以保持背景对噪声和离群值的鲁棒性。实验结果表明,所提模型在复杂场景中能较准确检测出运动目标。

Spatiotemporal RPCA algorithm for moving target detection in complex scene

Using robust principal component analysis(RPCA)is easy to misjudge the dynamic background as a foreground moving target under complex dynamic background,deteriorating its performance on moving target detection.To solve this problem,a spatiotemporal low-rank RPCA algorithm based on non-convex weighted kernel norm was proposed.The non-convex weighted nuclear norm was used to replace the traditional nuclear norm for low rank constraint.The Laplace matrix of the space-time graph was obtained through the Laplace feature mapping of the observation matrix.The obtained space-time graph Laplace matrix was embedded into the low-rank background matrix to maintain the robustness of the background against noise and out-liers.Experimental results show that the proposed model can accurately detect the moving target in complex scenes.