图3.4:行为动作的起止点目标回归曲线。在测试阶段,当起始点(终止点)的回归曲线到达局部峰值时,可以定位为行为动作的起始(结束)位置。 总结和展望 由于行为识别技术在智能监控、人机交互、视频序列理解、医疗健康等众多领域扮演着越来越重要的角色,研究人员正使出“洪荒之力”提高行为识别技术的准确度。说不定在不久的某一天,你家门口真会出现一个能读懂你的行为、和你“心有灵犀”的机器人,对于这一幕,你是不是和我们一样充满期待? [1] https://movie.douban.com/subject/25757903/ [2] Gunnar Johansson. Visual perception of biological motion and a model for it is analysis. Perception and Psychophysics 14(2), pp 201–211, 1973. [3] Alejandro Newell, Kaiyu Yang, Jia Deng. Stacked Hourglass Networks for Human Pose Estimation, In ECCV, 2016. [4] Zhe Cao, Tomas Simon, Shih-En Wei, Yaser Sheikh. Realtime Multi-person 2D Pose Estimation using Part Affinity Fields. arXiv preprint arXiv:1611.08050, 2016. [5] [6] CVPR2011 Tutorial on Human Activity Recognition.
[7] Wentao Zhu, Cuiling Lan, Junliang Xing, Wenjun Zeng, Yanghao Li, Li Shen, Xiaohui Xie. Co-Occurrence Feature Learning for Skeleton Based Action Recognition Using Regularized Deep LSTM Networks. In AAAI, 2016. [8] Sijie Song, Cuiling Lan, Junliang Xing, Wenjun Zeng, Jiaying Liu. An End-to-End Spatio-Temporal Attention Model for Human Action Recognition from Skeleton Data. Accepted by AAAI, 2017. [9] Yanghao Li, Cuiling Lan, Junliang Xing, Wenjun Zeng, Chunfeng Yuan, Jiaying Liu. Online Human Action Detection Using Joint Classification-Regression Recurrent Neural Networks. In ECCV, 2016. 作者简介
兰翠玲博士,微软亚洲研究院副研究员,从事计算机视觉,信号处理方面的研究。她的研究兴趣包括行为识别、姿态估计、深度学习、视频分析、视频压缩和通信等,并在多个顶级会议,期刊上发表了近20篇论文,如AAAI, ECCV, TCSVT等。 你也许还想看: 感谢你关注“微软研究院AI头条”,我们期待你的留言和投稿,共建交流平台。来稿请寄:[email protected]。 微软小冰进驻微软研究院微信啦!快去主页和她聊聊天吧。 (责任编辑:本港台直播) |