A Novel Safety Helmet Wear Detection Method based on Improved RFBnet. (August 2020)

• Record Type: Journal Article
• Authors: Chen, Liu; Xue, Zhishuang; Liu, Xiaofang; Chen, Mingju; Huang, Hui
• Is Part Of: Journal of physics. Volume 1617(2020)
• Publication Date: 2020-08
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1617/1/012035
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: In order to solve the problems of error detection and missed detection in complex environments, a helmet wear detection method based on dense connection method and CIou loss is proposed. This method is based on RFBnet-512 neural network, and improves the bankbone network of RFBnet-512 with dense connection to enhance the bankbone network's ability to extract features. At the same time, in order to avoid errors caused by inconsistent loss function and evaluation index, CIou loss is used to replace the original SmoothL1 loss of RFBnet-512 network, making the calculation of regression loss more close to the evaluation index. The experiment shows that compared with the original RFBnet-512 network, the detection accuracy is improved by 0.018, and the detection speed is significantly improved.