摘要
提出了一种基于时空关系和多观察值的三层隐马尔科夫扩展模型识别复杂交互活动的方法。根据多目标交互活动具有分层的性质和目标之间的时空关系,给出了提取3个粒度(整体,双人,单人)行为特征的方法。同时提出与之对应的多观测值三层隐马尔科夫扩展模型。实验结果表明:将新的特征提取方法和新的模型应用于复杂交互行为识别能得到较高的识别准确率和较好的鲁棒性。
A complex interactive activity recognition approach with spatial-temporal relation and an extended Hidden Markov Model with multi-observations and multi-layers is presented here. Interactive activities involving multi-objects are naturally hierarchical and related with spatial-temporal relationship. The multi-granularity features （group, two persons and single person） are used. A new model, Multi-observations Three-Layers Hidden Markov Model （MTHMM）, corresponding to these features is put forward. The experiments show that the new feature extraction method and the new model have a good performance and a fair robustness in complex interactive activity recognition.
出处
《吉林大学学报（工学版）》
EI
CAS
CSCD
北大核心
2014年第2期421-426,共6页
Journal of Jilin University:Engineering and Technology Edition
基金
国家自然科学基金项目(61133011
61303132
61103091)
吉林省科技发展计划项目(20140101201JC
201201131)
关键词
人工智能
时空关系
交互行为
三层多观察隐马尔科夫模型
三层特征
artificial intelligence；spatial-temporal relation；interactive activity；hidden Markov model with multi-observations and three-layers；features with three layers
