基于光流法的深度学习在工业运动检测的应用

Deep Learning on Optical Flow in Industrial Motion Detection Application

工业现场中常需要根据生产过程中目标的运动或变化状态进行控制、联锁或报警。通过视频图像的运动检测识别技术,可有效地代替人工作业,提高生产自动化水平。文中利用光流法成熟有效的运动特征提取能力,以及深度学习技术在图像分类识别上的突出表现,将两者相结合,利用AlexNet网络对可视化光流处理后的工业对象视频图像进行迁移学习的训练,取得了良好的识别效果,并成功应用在生产环境中。

In industrial field,it is often necessary to control,interlock or alarm according to the moving or changing state of the target in the production process. Video image motion detection and recognition technology can effectively replace manual work and improve the level of production automation. In this paper,using the mature and effective motion feature extraction ability of optical flow method and the outstanding performance of deep learning technology in image classification and recognition,combining the two methods,AlexNet network is used to train the transfer learning of the video images of industrial objects after visual optical flow processing,and good recognition effect is achieved,and successfully applied in production environment.