基于LSTM神经网络的手势分割  

Gesture Segmentation Based on LSTM Neural Network

作　　者：胡跃辉 钟纪权 易小斌 张涛 HU Yuehui;ZHONG Jiquan;YI Xiaobin;ZHANG Tao(Hefei University of Technology,Hefei,Anhui Province,230009 China;Hefei Xiliu Photoelectric Technology Co.,Ltd.,Hefei,Anhui Province,230009 China)

机构地区：[1]合肥工业大学,安徽合肥230009 [2]合肥溪流光电科技有限公司,安徽合肥230009

出　　处：《科技创新导报》2022年第24期85-89,共5页Science and Technology Innovation Herald

摘　　要：本文针对手势识别系统中基于阈值类的手势分割算法忽略了手势信号在时间上的相关性,高度依赖经验阈值等问题,提出了一种基于长短期记忆单元(LSTM)的网络结构应用于手势分割,将手势信号看成一段时间相关序列,利用LSTM神经网络的特点,对手势信号的动态信息加以学习,联合起过去信息,实现对当前帧的判决。实验结果表明,所提出的算法模型在准确率、可靠性方面优于传统方法,无需设定经验阈值即可快速实现手势信号的自动分割。Aiming at solving the problem that the threshold based gesture segmentation algorithm in gesture recog-nition system ignores the time dependence of gesture signals and highly relies on experience threshold,this paper proposes a network structure based on Long Short Memory Unit(LSTM)for gesture segmentation.The gesture sig-nal is regarded as a time related sequence,and the dynamic information of the gesture signal is learned by using the characteristics of LSTM neural network,which combines the past information to realize the decision of the current frame.The experimental results show that the proposed algorithm model is superior to the traditional methods in ac-curacy and reliability,and the automatic segmentation of gesture signals can be realized quickly without setting an empirical threshold.

关 键 词：手势分割 阈值法 长短期记忆 时间序列 手势识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]