一种基于证据推理的视频运动目标融合检测算法

为了解决复杂场景下传统的运动目标检测问题, 利用证据推理—–谨慎有序加权平均方法(COWA-ER), 提出一种综合使用混合高斯、均值滤波和码本的多方法融合的检测方法. 该融合检测算法以上述3 种检测方法为准则建立一个多准则决策框架, 通过双阈值检测法来表征检测过程中的不确定性, 最终利用COWA-ER 方法进行决策级融合, 实现多种方法的优势互补. 实验表明, 所提出的融合检测算法具有更理想的目标检测效果, 能有效应对诸如阴影及光照突变等问题对检测性能的影响.

A fusion approach for video moving object detection based on evidential reasoning

To handle the problems encountered in traditional moving object detecting algorithms under complex scenarios, by comprehensively using the Gaussian mixture model(GMM), mean filter and codebook(CB), a fusion-based detection approach is proposed based on cautious ordered weighted averaging with evidential reasoning(COWA-ER). In the proposed approach, a multi-criteria decision-making framework is established, where the three detection algorithms are used as multiple criteria. The double threshold method is used to model the uncertainty in the detection. The decision-level fusion for detection is finally accomplished by using COWA-ER, where complementary advantages of the three algorithms can be fully used. Experimental results show that the proposed approach can achieve better performance for the detection, and effectively deal with the influence of detection performance caused by shadow and light change.