基于改进LLE算法的手势特征提取研究  

Research on Gesture Feature Extraction Based on Improved LLE Algorithm

作　　者：张翔 席奇 刘承启[2] ZHANG Xiang;XI Qi;LIU Cheng-qi(School of Information Engineering,Jingdezhen University,Jingdezhen Jiangxi 333000,China;Nanchang University,Nanchang Jiangxi 330029,China)

机构地区：[1]景德镇学院信息工程学院,江西景德镇333000 [2]南昌大学,江西南昌330029

出　　处：《计算机仿真》2021年第12期142-145,159,共5页Computer Simulation

基　　金：江西省教育厅科技项目(GJJ181132)。

摘　　要：在人机交互领域中,基于视觉的手势特征提取成为研究的热点,但手势存在较大范围的变化,很难实现对手势的有效分类。研究了一种基于LLE改进算法的手势特征提取方法。先将手势特征数据中的某个数据点与邻近数据点组成局部线性关系,对重构误差进行拉格朗日乘子算法优化处理,求出新的局部重建权值矩阵,为了使局部线性关系能够满足低维度空间,通过求解映射矩阵的方法,将手势特征样本数据的目标特征空间映射到低维度空间中。采用稀疏观察手势描述法对手势特征进行提取,根据手势参数对手势轨迹数据进行归一化处理,为了提高手势特征提取的实时性,采用支持向量分类器的方法将手势从难以分类的空间映射到高维度的手势空间中。实验结果表明,所提方法对手势的查准率和召回率较高,泛化性较好,即使在样本数据很少的情况下,也具有较好的识别效果。In the field of human-computer interaction, gesture feature extraction based on vision has become a re-search hotspot, however, there are a large range of changes in gestures, so it is difficult to realize the classification ofgestures. A gesture feature extraction method based on improved LLE algorithm was studied. Firstly, a local linear re-lationship was formed between a certain data point and adjacent data points in gesture feature data, and the recon-struction error was optimized by Lagrange multiplier algorithm to obtain a new local reconstruction weight matrix. Inorder to make the local linear relationship meet the low dimensional space, the target feature space of gesture featuredata was mapped to the target feature space by solving the mapping matrix in low dimensional space. Secondly, sparseobservation gesture description method was used to extract gesture features, and gesture trajectory data were normal-ized according to gesture parameters. In order to improve the real-time performance of gesture feature extraction, sup-port vector classifier was used to map gesture from the difficult space to high-dimensional gesture space. Experimentalresults show that the proposed method has high precision and recall rate, good generalization, and good recognitioneffect even in the case of few sample data.

关 键 词：手势特征提取 局部重建权值矩阵 映射矩阵 稀疏观察 归一化处理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]