中文摘要：

针对目前人体行为检测系统,大多是基于对图像中目标进行短时间检测的情况,提出了一种基于条件随机场模型和区域划分的异常行为检测方法。该方法利用GMM模型提取目标,去除阴影后,获得目标轨迹坐标序列,结合分类后的场景,利用划分区域和轨迹,对CRF进行建模,之后利用构建好的模型,计算待测视频属于不同进行异常行为的概率检测。与采用HMM方法相比,实验表明,此采用方法充分利用了长距离特征,提高了检测率。

英文摘要：

Considering nowadays that the human behavioral detection system is mostly based on the short description of target image,a method of abnormal detection based on CRF（conditional random fields） and background zoning is proposed.The method uses GMMs model to extract target,get the coordinates of target,and combines the background scene classification.And the coordinates and the zoned background are used to build the CRFs model,then the CRFs model is used to calculate the probability belonging to different kinds of abnormal behavior,and classify the abnormal behavior in test video.This method is compared with the method with HMMs model.Experiment indicates that this method with CRFs could make full use of the Long-range features,and thus improve the detection rate.