摘要
针对基于传统社会力模型的人群异常行为检测算法忽视了场景中各个区域信息差异性的问题,考虑到时空LBP序列谱特征的计算简单性和区域代表性,提出了一种基于时空LBP加权社会力模型的人群异常行为检测算法,将时空LBP序列谱特征所包含的时域特性和区域信息融入社会力模型,使得社会力模型更为精确地对人群行为进行建模。实验证明,与传统算法相比,改进后的算法在异常行为的查准率与查全率上有很大的提高。
In view of the situation that traditional anomaly detection algorithms using social force model usually ignore the differences between different regions of the scene, a method of abnormal crowd behavior detection is proposed based on spatiotemporal LBP-weighted social force model. This algorithm integrates the time domain characteristics and regional information contained in the spectral characteristics of spatiotemporal LBP sequence into the social force model ,enabling the model of the crowd more accurate. Experiments show that recall and precision of the proposed algorithm are greatly improved.
出处
《电视技术》
北大核心
2012年第21期145-148,共4页
Video Engineering
基金
上海市科委科技攻关项目(11231203102)
国家自然科学基金项目(61102099)
关键词
光流
局部二值模式
社会力模型
异常检测
optical flow
LBP
social force model
abnormal crowd behavior detection