Abstract: For the problem of the dynamic-free action recognition being susceptible to noise，corruption and occlusion，this paper proposes a robust action classification approach based on a sparse representation.A testing action is treated as a sparse linear combination of all training actions which is extended to contain a error term，and its sparsest representation is computed by minimizing the [l1] norm of both coefficients and error.The testing action is classified by minimizing the residual.The experiments are evaluated on the Weizmann robustness test sequences.The results demonstrate that the algorithm developed is robust to noise，corruption and occlusion.

Key words: action classification, sparse representation, [l1]minimization, compressive sensing

摘要： 针对无动态性的动作识别中易受噪声、干扰和遮挡等影响的问题，提出了一种基于稀疏表示的鲁棒的动作分类方法。对要测试的动作表示成所有训练动作的稀疏线性组合，并扩展该稀疏表示方程使其包含错误项，通过对系数和错误项的l1范数最小化算法来求解其最稀疏的表示，根据所得的稀疏解基于最小剩余量进行分类。并在Weizmann鲁棒性测试序列上进行了评价，实验结果表明该算法对噪声、干扰和部分遮挡具有较好的鲁棒性。

关键词: 动作分类, 稀疏表示, l1最小化, 压缩传感