摘要
提出了基于奇异值分解(Singular Value Decomposition,SVD)特征矩阵压缩和隐Markov模型(Hidden Markov Model,HMM)的动态手势识别方法。该方法通过SVD对特征矩阵进行时间维度的压缩,然后通过HMM的方法对提取的动态手势进行识别。通过对特征矩阵压缩可以显著地减少训练HMM的迭代计算量,提高模型的训练效率。采用Leap Motion体感控制器追踪并提取自定义的10个阿拉伯数字的动态手势特征。实验验证结果表明,该方法对这些动态手势在当前有限样本条件下的总识别率均在96%以上。
In the paper, a dynamic gesture recognition method based on SVD feature compression and HMM is proposed. The method uses SVD to compress the feature matrix, and then implements the dynamic gesture recognition by training the HMM gesture model. The feature ma- trix compression by SVD can significantly reduce the computational complexity of the HMM model and improve the training efficiency. This pa- per uses Leap Motion controller to track and collect the customized dynamic gesture features of ten Arabia figures. Experimental result shows that this method can classify the samples with a total accuracy over 96%.
出处
《信息技术与网络安全》
2018年第1期106-110,共5页
Information Technology and Network Security
