Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (12): 181-187.DOI: 10.3778/j.issn.1002-8331.1812-0002

Previous Articles     Next Articles

Automatic Induced Maintenance Process Interaction Design Based on Faster R-CNN

LUO Youwen, WANG Wei, QU Jue   

  1. Air and Missile Defense College, Air Force Engineering University, Xi’an 710051, China
  • Online:2019-06-15 Published:2019-06-13

基于Faster R-CNN的诱导维修自动交互设计

罗又文,王  崴,瞿  珏   

  1. 空军工程大学 防空反导学院,西安 710051

Abstract: With the development of augmented reality technology in the field of machinery, more and more examples have proved the superiority of AR in improving the operational efficiency in industrial maintenance. In order to improve the efficiency of the induced maintenance operation process, an automatic interaction of process recognition based on Faster Regional Convolutional Neural Network(Faster R-CNN) is proposed for the problem that the traditional augmented reality maintenance system cannot sense and judge the maintenance status. The method is based on Faster R-CNN to establish a deep neural network model for part recognition and further fine-tuning by means of back propagation. By identifying the type and number of the parts, the feedback is given to the system to trigger the corresponding operation steps without additional interaction by the user operating. The experimental results show that the recognition rate of repair parts based on deep neural network can reach 95%, and the average recognition speed is 300 ms per frame, which meets the accuracy and interactivity requirements of AR-induced maintenance system.

Key words: augmented reality, deep learning, Faster Regional Convolutional Neural Network(Faster R-CNN), induced maintenance, automatic interaction

摘要: 随着增强现实技术在机械领域的发展,已经有越来越多的例子证明了AR在工业维修方面提高操作效率的优越性。为了提高诱导维修操作过程的效率,针对传统的增强现实维修系统不能对维修状态进行感知和判断的问题,提出了一种基于快速区域卷积神经网络(Faster R-CNN)的进程识别自动交互方法。该方法基于Faster R-CNN建立零件识别的深度神经网络模型并利用反向传播进一步微调,通过对零件的识别输出零件的类型和编号,反馈给系统触发相应的操作步骤,无需用户进行另外的交互操作。实验结果表明,基于深度神经网络的维修零件识别率可达95%,平均识别速度为每帧300 ms,满足AR诱导维修系统的精度和交互性要求。

关键词: 增强现实, 深度学习, 快速区域卷积神经网络(Faster R-CNN), 诱导维修, 自动交互