人体下肢应激微反应自动识别

发布时间：2018-07-05 18:44

本文选题：人体下肢 + 应激微反应　；参考：《光学精密工程》2017年11期

【摘要】：由于现有的动作识别方法不能直接用于人体微反应动作识别,本文基于人体下肢微反应动作特点,构建了一种时空金字塔韦伯局部描述子并设计了基于字典学习的人体下肢微反应自动识别算法。该方法利用时空金字塔韦伯局部描述子提取每一类人体下肢微反应动作特征,使用主成分分析法对特征降维;然后,建立每一类动作子字典并将子字典串联形成总的动作字典;最后,通过实验分析了金字塔级数L,降维后每类动作特征维数d_(PCA),每类动作子字典原子个数n_(Atom),以及稀疏阈值C等参数对识别结果的影响,并确定最优参数值L=3,d_(PCA)=30,n_(Atom)=40,C=10。实验结果表明,提出的算法对10种人体下肢微反应动作的识别率均在0.83~0.91之间,平均识别率达到0.86,高于其他动作识别算法。设计的算法更适用于人体下肢微反应动作分类,并可有效提高分类识别率。
[Abstract]:Because the existing motion recognition methods can not be directly used in human micro-reaction action recognition, this paper bases on the characteristics of human lower extremity micro-reaction action. A Webb local descriptor based on temporal and spatial pyramids is constructed and an automatic recognition algorithm for human lower limb microreactions based on dictionary learning is designed. In this method, the local descriptors of time and space pyramidal Weber are used to extract each kind of human lower extremity micro-reaction action feature, and the principal component analysis is used to reduce the dimension of the feature. Then, the dictionary of each kind of action sub-is established and the sub-dictionary is connected in series to form the total action dictionary. Finally, the effects of parameters such as pyramid series L, characteristic dimension d _ (PCA) of each action after dimensionality reduction, the number of dictionary atoms n _ (Atom), and sparse threshold C of each action subclass on the identification results are analyzed, and the optimal parameter L _ (3) D _ (PCA) is determined. The experimental results show that the recognition rate of the proposed algorithm for 10 kinds of human lower extremity micro-reaction actions is between 0.83 and 0.91, and the average recognition rate is 0.86, which is higher than that of other algorithms. The proposed algorithm is more suitable for the classification of human lower extremity micro-reaction and can effectively improve the classification recognition rate.
【作者单位】：山东警察学院;山东大学高效洁净机械制造教育部重点实验室;山东行政学院;
【基金】：国家自然科学基金资助项目(No.51405194) 山东省自然科学基金资助项目(No.ZR2016EEP13,No.ZR2013CQ025) 山东省高等学校科技计划资助项目(No.J16LN16)
【分类号】：TP391.41

【相似文献】