Eye State Detection Algorithm Based on Shallow Dual-scale Siamese CNN for manufacturing workshop workers

Eye State Detection (ESD) is an important means of safety risk detection during the interaction between humans and computers in a workshop setting. The deep network model incurs a significant computational cost, making it unsuitable for deployment on computing devices that have weak computing power. This work created a shallow dual-scale Siamese-based ESD model. First, the shallow dual-scale Siamese network is used to extract eye image features, and the high-dimensional features of the Siamese channels are subjected to feature fusion and principal component analysis. Finally, this network is inputted into the Support Vector Machine(SVM) for training and testing. Experiments on the Open and Close Eyes(OACE) and Human Eyes(HE) public datasets show that the detection speed and detection performance comprehensive indicators of the detection algorithm are better than the other seven deep baseline models, and it has better robustness on color and grayscale images with different light intensities and color depths.