Simple Industrial Cutting Machine Safety System Based on Computer Vision

Safety devices are often not fitted to many simple panel-cutting machines due to cost. Designing and installing low-cost safety devices on the simple panel-cutting machine improves the safety of operators. We propose a computer vision-based safety system with a camera to capture images of the operator and machine from above. Each captured image is analyzed by a neural network. If the operator's hands are too close to the cutting tool, the safety system sends a warning signal and stops the machine. The neural network used in the vision system is a ten-layer convolutional neural network. In addition to the input and output layers, there are five convolutional layers in the middle, followed by three fully connected layers. The input image undergoes five convolution operations to obtain feature maps of different granularity. With three fully connected layers, a combination of features is used to determine the positions of the operator's hands as a guideline for stopping the machine. To improve the generalization of the model, we collect thousands of training data. The experimental results show that the model has a low bias, low variance, and high accuracy characteristics. The accuracy of the verification data is over 98%.