• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2011, Vol. 33 ›› Issue (5): 74-79.

• 论文 • 上一篇    下一篇

粒子群优化神经网络在动态手势识别中的应用74

李文生,姚琼,邓春健   

  1. (电子科技大学中山学院计算机工程系,广东 中山 528402)
  • 收稿日期:2010-11-20 修回日期:2011-02-18 出版日期:2011-05-25 发布日期:2011-05-25
  • 作者简介:李文生(1966),男,湖南郴州人,硕士,副教授,研究方向为数字信号处理。姚琼(1977),女,江西南昌人,硕士,讲师,研究方向为数字图像处理和模式识别。邓春健(1980),男,广东韶关人,博士,副教授,研究方向为信息显示技术和通信技术。
  • 基金资助:

    广东省自然科学基金资助项目(8152840301000009);广东省科技计划资助项目(2009B030803031)

Application of the BP Neural Network Based on PSO in Dynamic Gesture Recognition

LI Wensheng,YAO Qiong,DENG Chunjian   

  1. (Department of Computer Engineering,Zhongshan Institute,University of
    Electronic Science and Technology of China,Zhongshan 528402,China)
  • Received:2010-11-20 Revised:2011-02-18 Online:2011-05-25 Published:2011-05-25

摘要:

为了提高动态手势学习训练速度和识别准确率,本文提出一种基于粒子群优化BP神经网络的动态手势识别方法。首先基于自然人机交互需要,定义一套基于机器视觉的动态手势模型;在获取指尖运动轨迹的基础上,提取动态手势的特征向量作为神经网络的输入;利用改进的PSO算法训练BP神经网络,得到神经网络的权值和阈值;最后利用训练过的神经网络识别基于机器视觉的动态手势。测试结果表明:改进的PSO算法能够提高神经网络训练速度和精度,进而提高动态手势识别准确率。

关键词: 机器视觉, BP神经网络, 动态手势识别, 粒子群

Abstract:

In order to improve the training speed and identification accuracy of dynamic gesture, a method of gesture recognition based on the particle swarm optimization(PSO) BP neural network is  put forward. First, a set of dynamic gestures is defined for HumanMachine Interaction (HMI). The engenvectors vectors of dynamic gestures are extracted as the input of the BP neural network on the basis of obtaining the trajectories of moving fingertips. An improved PSO algorithm is used to train the BP neural network and get the weights/thresholds of the network. Finally, the gestures based on machine vision are  recognized through the trained BP neural network. The experimental results show that the proposed PSO algorithm can enhance the speed and precision of network training, and improve the accuracy of dynamic gesture recognition.

Key words: machine vision;BP neural network;dynamic gesture recognition;particle swarm optimization