Real-time Safety Helmet Detection System based on Improved SSD

Abstract

The detection of the safety helmet is difficulties due to the ariable ighting, weather changes and complex background. We proposed a deep learning detection method to detect safety helmet to solve the problems of low accuracy and poor robustness of traditional detection methods. This method is based on the SSD (Single Shot MultiBox Detector) object detection and improved the network. First, we used the fusion of multi-layer to consideration of shallow low sematic information and deep semantic information, which improves the sensitivity of the network to small target detection. Second, we proposed lightweight network structure of compresses the network, reducing the amount of parameters and calculations of the model. Third, we made safety helmet datasets to train and test the improved network model, and the model is compared with the original SSD. The results show that the detection accuracy of the model is 86.75%, which is similar to SSD, but the detection speed has been improved significantly, which is 295% higher than SSD, up to 83 frame/s. Experiments show that the improved network model can significantly improve the detection speed while ensuring the detection accuracy and meet the real-time detection requirements.