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dPCA，每类动作子字典原子个数nAtom，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，dPCA=30，nAtom=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。

Because the existing motion recognition method couldnot be directly used in human micro-expression action recognition. A spatio-temporal pyramid Weber Local Descriptor (STPWLD) was constructed and an automatic recognition algorithm of human lower limb micro-expression action based on dictionary learning according to characterize human lower limb micro-expression action was designed. With the method， the features of human lower limb micro-expression action was extracted by the STPWLD. Then， the dimensions of STPWLD feature were reduced by the principal component analysis. Furthermore， the sub-dictionaries of human lower limb micro-expressions action was established and these sub-dictionaries were connected in series to construct a general action dictionary. Finally， the influence of the parameters of the algorithm on the recognition results was analyzed， and the optimal value of these parameters was determined. It shows that the optimal value of pyramid scales is 3， the optimal feature dimension of each action after dimension reduction is 30， the optimal number of atoms in each action dictionary is 40 and the optimal value of sparse threshold is 10. The experimental results indicate that the recognition rates of the proposed algorithm for 10 kinds of human lower limb micro-expression actions are all between 0.83~0.91， and the average recognition rate is 0.86， higher than that of other algorithms. The algorithm is suitable for the classification of human lower limb micro-expression actions and improves the classification recognition rate effectively.