基于自编码的中国手语识别研究  

Chinese sign language recognition research based on auto-encoder

作　　者：林鹏程[1,2] 林培杰[1,2] 程树英[1,2] 

机构地区：[1]福州大学物理与信息工程学院,福建福州350116 [2]福州大学微纳器件与太阳能电池研究所,福建福州350116

出　　处：《微型机与应用》2017年第13期59-61,65,共4页Microcomputer & Its Applications

摘　　要：针对动态手语上下文联系强的特点,采用LSTM(Long Short-Term Memory)循环神经网络用于识别,同时,利用自编码实现动态手语的无监督学习。该方法将数据手套作为采集设备,在获取手语信息后,通过编码器、解码器生成手语的重构矢量。在实验过程中,将采集的数据集进行模型的训练,并与监督学习的结果进行比较。实验结果表明,该方法能够有效进行手语识别的无监督学习。Because dynamic sign language has strong context-sensitive feature,we use recurrent neural networks based on LSTM to recognise it. At the same time, auto-encoder is used to realize unsupervised learning with data glove as the collecting device. After obtaining the information of sign language,the data is used to reconstruct vector of sign language through encoder and decoder. During the experiment,the collected data is used to train the model and compare with the result of supervised learning. Experimental results demonstrate the effectiveness of the proposed auto-encoder on unsupervised learning sign language.

关 键 词：长短期记忆 自编码 数据手套 手语识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]