摘要
针对目前行为识别方法的不足,提出一种基于人体3D骨架和多CRF模型(MCRF)的行为识别方法.3D骨架数据量少且保留了行为关键信息的优点,并具有融合多特征和上下文信息的优势.为此,首先基于3D骨架将人体动作划分为全局运动、手臂运动和腿部运动,通过对动作序列进行多类特征提取,形成多类特征集;然后利用CRF模型对每一特征集建模,再融合所有的CRF模型,得到MCRF模型;最后利用MCRF模型进行行为识别.实验结果表明,该方法具有较高检测率.
Considering the disadvantages of the traditional human activity recognition system,a human activity recognition system using an MCRF model and 3D skeletons was proposed.Its 3D skeleton data has less data and retains the key information,and the MCRF model has the advantage of being able to combine more features and utilizing adaptive contextual information.First,human activity was divided into global activity,arm activity,and leg activity.Several feature subsets were formed through more feature extraction.Then,CRF models were used on each feature subset to generate CRF units.Finally,all the CRF units were combined to produce the MCRF model which was utilized to recognize human activity.The experimental results indicate that the proposed method can improve detection accuracy.
基金
国家自然科学基金(61071173)资助
关键词
行为识别
3D骨架
MCRF
特征提取
human activity recognition
3D skeleton
MCRF
feature extraction
