Abstract: Helmet has been widely used in industrial production. In order to prevent accidents and ensure production safety, it is quite important to set up an automatic detecting and warning system. This paper focuses on the recognition algorithm on helmet. By using skin color detection to locate the face region, the region images above the face can be obtained. Moreover, Hu moments are taken as image feature vectors to compare neural network and SVM as classification models. Experimental results show that SVM can attain a better recognition on helmet and provides a stronger support and practical guidance for the realization of intelligent monitoring system.