基于时空信息的时序动作检测方法研究
|
摘要
本文提出了一个深度时空信息网络.加入了反映动作时空信息的光流来获取时序信息,通过3D卷积网络检测结果,得到视频中动作发生的候选区域及其动作分类.在此基础上,本文通过构建动作状态检测网络,对得到的候选区域进行修补,从而可以得到更为精确的动作发生的时间区域.实验结果表明,相对于现有的方法,本文的方法有效地提高了时序动作区域的定位精度.
This paper proposes a deep space-time information network(DSTIN)for the detection of temporal action regions.our method added optical flow information as an input to get the temporal information,uses 3D convolutional networks to get candidate regions and classify the actions.Then our method constructed and trained a specialized 3D convolutional network to detect the state of the candidate regions and perform modification on those regions.Experiment result shows that our method can effectively improve the accuracy of candidate regions for temporal action detection than the existing methods.
This paper proposes a deep space-time information network(DSTIN)for the detection of temporal action regions.our method added optical flow information as an input to get the temporal information,uses 3D convolutional networks to get candidate regions and classify the actions.Then our method constructed and trained a specialized 3D convolutional network to detect the state of the candidate regions and perform modification on those regions.Experiment result shows that our method can effectively improve the accuracy of candidate regions for temporal action detection than the existing methods.
引文
[1]陈颖鸣,陈树越,张显亭.智能视频监控中异常行为识别研究[J].微电子学与计算机,2010,27(11):102-105.
[2]付朝霞,王黎明.基于时空兴趣点的人体行为识别[J].微电子学与计算机,2013,30(8):28-30.
[3] Shou Z,Chan J,Zareian A,et al.Cdc:Convolutionalde-convolutional networks for precise temporal action localization in untrimmed videos[C]∥2017IEEE Conference on Computer Vision and Pattern Recognition(CVPR).2017:1417-1426.
[4] SHOU Z,WANG D,CHANG S F.Temporal action localization in untrimmed videos via multi-stage cnns[C]∥2016IEEE Conference on Computer Vision and Pattern Recognition(CVPR).2016:1049-1058.
[5] GAO J,YANG Z,SUN C,et al.Turn tap:Temporal unit regression network for temporal action proposals[C]∥2017IEEE International Conference on Computer Vision(ICCV).2017:3648-3656.
[6]HE K,ZHANG X,REN S,et al.Deep residual learning for image recognition[C]∥2016IEEE Conference on Computer Vision and Pattern Recognition(CVPR).2016:770-778.
[7] YEUNG S,RUSSAKOVSKY O,MORI G,et al.End-to-end learning of action detection from frame glimpses in videos[C]∥2016IEEE Conference on Computer Vision and Pattern Recognition(CVPR).2016:2678-2687.
[2]付朝霞,王黎明.基于时空兴趣点的人体行为识别[J].微电子学与计算机,2013,30(8):28-30.
[3] Shou Z,Chan J,Zareian A,et al.Cdc:Convolutionalde-convolutional networks for precise temporal action localization in untrimmed videos[C]∥2017IEEE Conference on Computer Vision and Pattern Recognition(CVPR).2017:1417-1426.
[4] SHOU Z,WANG D,CHANG S F.Temporal action localization in untrimmed videos via multi-stage cnns[C]∥2016IEEE Conference on Computer Vision and Pattern Recognition(CVPR).2016:1049-1058.
[5] GAO J,YANG Z,SUN C,et al.Turn tap:Temporal unit regression network for temporal action proposals[C]∥2017IEEE International Conference on Computer Vision(ICCV).2017:3648-3656.
[6]HE K,ZHANG X,REN S,et al.Deep residual learning for image recognition[C]∥2016IEEE Conference on Computer Vision and Pattern Recognition(CVPR).2016:770-778.
[7] YEUNG S,RUSSAKOVSKY O,MORI G,et al.End-to-end learning of action detection from frame glimpses in videos[C]∥2016IEEE Conference on Computer Vision and Pattern Recognition(CVPR).2016:2678-2687.