Virtual Fence Approach for Safety Monitoring of CNC Machine Tool with Dual Regions of Interest

Abstract

This study is about machine vision-based virtual fence technology for CNC machine tools. The proposed smart system is composed of cameras and industrial personal computers, and the object of concern is the safety door. First, the operation mode of the camera is in real-time, and the industrial personal computer is responsible for image storage, analysis, and decision-making. In the total field of view, key regions of interest are predefined by experts, including the safety door and the work area, which represent the behavior of the workers and the operating status of the machine, respectively. However, since the contrast between the door opening and closing is significant, a background subtraction method is employed to monitor the status of the safety door. Then, the work area adjacent to the safety door has the ability to detect foreign intrusion. Its properties are different from the safety door in contrast, and the temporal difference learning method is more suitable for solving this problem. The experimental results show that the proposed method can accurately detect the state of the safety door and the intrusion of foreign objects. Furthermore, the proposed method still has reliable performance within tolerable error after reducing the detection frequency, which also confirms that the trade-off problem of computational cost and accuracy can be taken into account at the current stage of this study. Future research will focus on determining regions of interest and keyframes, as well as performance testing for object detection and semantic segmentation.