摘要
针对现阶段人群行为分析的特征提取效果不佳,人群行为分析结果达不到视频分析的要求。提出一种基于人群群组级别的动量特征,分别表示人群的集体性、稳定性和冲突性,然后将三组人群群组动量特征输入至卷积神经网络进行训练,最后在Violence数据集上进行人群行为分析实验。实验结果表明,提出的群组动量特征能够在群组级别表达出人群的基本特性,这些特性在人群行为分析中能够建立可识别较高的特征,在Violence数据集上的测试结果显示。提出的群组动量特征能够扩展到独立场景,对于任何场景的人群行为分析都能够获得鲁棒的基础动量特征,而采用卷积神经网络进行的训练和分类,能够提升人群行为分析的精确度。与传统特征及分类方法相比,在各种不同的独立场景中,将标注结果精度提升了13%左右,在视频场景人群行为分析中具有较强的实践意义。
Aim at the performance of crowed behavior analysis had a common feature extraction at the presentstage,the results of crowd behavior analysis c a n i reach the requirement of surveillance analysis, this articleproposes a motion feature based on the group level of c r o w d,and the motion feastability and conflict. Then, importing the three groups of motion features in the convolution neural network f rtraining,after training the algorithm getting the behavior labels for crowd. At last,som e contrast ecomparing on Violence dataset. Experimental results show that,the proposed group-level mthe most basic features for c r o w d, it will build discriminative features for recognition, and the behavior analysis result of Violence s how that group-level motion features can extend into independent-scene,for any scene it can obtain robust basic motion feature. A n d the convolution neural network for training and recognitioaccuracy fr crowd behavior analysis,compared with convention algorithms,the proposed algorithm independent scene,and improved the average accuracy of 13 % ,this research will do contribution for surveillancescene crowd behavior analysis.
作者
成金庚
计科峰
CHENG Jin- geng JI Ke- feng(Electronic Science and Technology School,National University of Defense Technology,Chang sha 410073,P . R . China)
出处
《科学技术与工程》
北大核心
2017年第14期79-85,共7页
Science Technology and Engineering
关键词
人群行为分析
视频场景
群组动量特征
深度卷积神经网络
crowd behavior analysis surveillance scene group-level motion features convolutionneura3network