摘要
姿态描述在行为识别、人机交互、异常检测等领域有着重要应用。根据姿态描述实时和准确性的要求,提出了结合原子体态以及原子体态之间组合规则的描述方法。采用人体骨架模型描述原子体态,对常用动作建立了骨架模型数据库,从一定程度上解决了单视环境下的遮挡问题。对训练样本,参照模型数据库完成编码,通过隐马尔科夫模型训练得到参数模型,并对训练得到的模型进行验证,以及测试。实验结果表明,该方法具有计算量小和准确率高的优点。
Posture description plays an important role in such fields as human behavior recognition, human-computer interaction, and abnormal events detection. Based on the requirements of real-time and exact posture description, a new method integrated with primary postures and combination rules of between the primary postures is proposed. With star-skeleton model for description of primary postures and by establishing star-skeleton database on normal human action, the overlapping problem in single vision condition is solved to a certain extent. The training samples are coded in accordance with the stat-skeleton database, while from HMM training the parametric models are generated. Test and simulation on the generated models indicate that the method proposed in this paper is of relatively low computation and fairly high accuracy.
出处
《通信技术》
2012年第12期91-94,共4页
Communications Technology
基金
国家自然科学基金(批准号:61071173)
关键词
姿态描述
骨架
隐马尔科夫
posture description
star-skeleton
HMM
