【摘要】 目标检测问题是视频监控系统中的关键问题,在监控场景复杂多变时,非参数方法是背景模型建模的重要方法,但传统的基于核密度带宽的非参数方法没有考虑像素之间的上下文关联性。本文提出建立目标检测问题的最大后验概率——马尔可夫随机场(MAP-MRF)框架,首先确定先验概率的能量函数,再确定条件概率及其能量函数,从而得到后验能量函数,采用蚁群算法迭代计算后验能量函数,获得像素标记的最优化结果即实现目标检测。在多个监控视频上的实验结果表明,本方法能大大提高复杂背景下的目标检测性能。更多还原

【Abstract】 It is important to detect objects accurately in visual surveillant systems. The non-parametric method is effective for detecting objects in complex surveillant scenes. But the traditional non-parametric method ignored the context of pixels. We present building the MAP-MRF framework of the object detecting problem. Firstly,the energy function of prior probability is defined,then,the energy function of conditional probability is acquired to get the posterior energy function. Consequently,the problem of detecting objects is defined as optimizing posterior energy function which is implemented by the ACS algorithm. Therefore,multiple objects are detected accurately via the optimal pixel labels. Experimental validation of the proposed tracking method is verified and presented on sequences.更多还原