中文摘要：

针对视频中异常目标行为特征的有效表示问题,提出一种基于动态感知模型的异常目标发现方法。对视频场景中的光滑、纹理、边沿区域建立动态感知模型,得到运动注意块作为候选检测位置,减少了对非感兴趣区域的冗余计算,再提取运动注意块的时空HNF特征使用稀疏编码算法训练生成字典。根据样本关于字典的重构误差是否超过预设阈值作为个体异常发现的判别标准。实验结果与测试数据库Ground Truth比较说明了该方法的有效性和实用性,且易于实现。

英文摘要：

Aiming at the troblem of representing behavior characteristics effectively of abnormal objects in video,this paper proposed an anomaly discovery method based on motion perception model. Firstly,it modeled smooth,texture and edge region in video scene by motion perception model,and extracted motion attention blocks as candidate position for detecting,reducing redundant computation for non-interested region in the algorithm. Sencondly,it extracted HNF feature of motion attention blocks in spatial-temporal dimensions as samples to produce a dictionary by sparse coding algorithm. It exploited reconstruction error of the sample based on the dictionary as discrimination criterion. Experiment shows that the proposed method is effective,practical and implemented easily.