兰州理工大学学报

兰州理工大学学报 ›› 2024, Vol. 50 ›› Issue (1): 96-103.

• 自动化技术与计算机技术 • 上一篇    下一篇

基于LSTM的智能手机3D手写识别

张乐1, 包广斌*2, 郭琳1, 武立3   

  1. 1.商洛学院 电子信息与电气工程学院, 陕西 商洛 726000;
    2.兰州理工大学 计算机与通信学院, 甘肃 兰州 730050;
    3.陕西省商洛市体育运动中心, 陕西 商洛 726000
  • 收稿日期:2022-10-10 出版日期:2024-02-28 发布日期:2024-03-04
  • 通讯作者: 包广斌(1975-),男,甘肃兰州人,博士,副教授. Email:baogb10@sina.com
  • 基金资助:
    陕西省教育厅专项科研计划项目(22JK0365),甘肃省自然科学基金(18JR3RA156),兰州市科技计划项目(2017-4-105)

3D handwriting recognition of smartphone based on LSTM

ZHANG Le1, BAO Guang-bin2, GUO Lin1, WU Li3   

  1. 1. Electronic Information and Electrical Engineering College, Shangluo University, Shangluo 726000, China;
    2. School of Computer and Communication, Lanzhou Univ. of Tech., Lanzhou 730050, China;
    3. Shangluo Sports Centre in Shaanxi Province, Shangluo 726000,China
  • Received:2022-10-10 Online:2024-02-28 Published:2024-03-04

PDF

64

摘要： 针对传统传感器需要在特定的空间区域内才能进行人机交互,极易受到外部环境因素干扰的问题,提出一种新的基于长短时记忆神经网络(LSTM)的智能手机3D空间手写识别方法,用于非特定三维空间中实现的人机交互.首先,利用智能手机内置三轴加速度传感器,采集手部运动数据,并将采集的数据进行预处理操作,构建3D手写识别数据集;然后,基于LSTM构建3D手写识别模型,并利用构建的数据集进行训练;最后,利用训练后的模型实现智能手机的3D手写分类识别.通过在本文自建的非依赖用户数据集上进行测试,实验结果表明,该识别方法可以实现86.4%的准确率,88.1%的召回率,88.4%的精准率和88.0%的F1分数.

关键词: 智能手机, 加速度传感器, 手写识别, LSTM

Abstract: Traditional sensors are prone to receive the interference of external environmental factors due to achieving human-machine interaction in specific spatial area. 3D handwriting recognition of smartphones based on the long short-term memory (LSTM) neural network is proposed, which can be used in human-machine interaction in non-specific 3D spaces. First, three-axis acceleration sensors of smartphones are used to collect data which perform pre-processing operations to construct a 3D handwriting recognition dataset. Then, the 3D handwriting recognition model based on LSTMis constructed and pre-trained by adopting the constructed datasets. Finally, the trained model is applied to implement 3D handwriting classification recognition for smartphones. By testing on a self-built non-dependent user dataset, experimental results show that the proposed model can achieve the accuracy rate of 86.4%, recall rate of 88.1%, precision rate of 88.4%, and F1 score of 88.0%.

Key words: smartphone, acceleration sensor, 3D handwriting recognition, LSTM

中图分类号: