摘要
提出了基于多示例学习法的人体行为识别方法。利用人体骨架模型,将人体主要关节的属性特征作为人体运动的几何特征,提出了一种基于行为几何特征的自适应行为分解算法,将行为分解为简单动作。把分解后的行为看作一个包,各个动作看作包中的各个示例,结合多示例学习法与Any Boost算法提出了多示例行为学习算法(MILBoost算法),通过在多示例框架下对每一类行为进行学习,得到强分类器用于未知行为包的识别。实验结果表明该方法与其他方法相比具有更高的识别精度,并且在有噪声或干扰的情况下具有很好的识别精度。
A novel method based multiple instance learning is proposed for human activity recognition.Based on the skeleton modelol human , a sequence major joints features are used as the geometriccharacteristics of human motion and a adaptive segmentation algorithm is introduced to decompose activityinto simple actions. The decomposed activity comprises labeled bags that are composed of unlabeledinstances comprising to action. MILBoost algorithm is presented by combine multiple instance learningliterature with the AnyBoost framework. Labeled behaviors are used to train a strong classifier which isused to predict the labels of unseen behavior bags. The experimental results show the effectiveness of theproposed method in comparison with other related works in the literature and can also tolerate noise andinterference conditions.
出处
《信息技术》
2016年第7期65-70,共6页
Information Technology
基金
国家自然科学基金项目(50808025)
中山市科技局工业攻关计划项目(2013A3FC0263)
关键词
行为识别
多示例学习
行为分解
包
activity recognition
multiple instance learning
activity decomposition
bag