Improved YOLOv3 Helmet Detection Algorithm
Da Yan, Liang Wang
Abstract
In order to solve the problem of low detection accuracy and slow speed of helmet wearing detection in video surveillance of construction site, YOLO-Helmet, a helmet wearing detection algorithm based on improved YOLOv3(You Only Look Once V3) is proposed. The network structure is improved on the basis of the YOLOv3 algorithm. Firstly, the size of the input image is increased. Secondly, using depthwise separable convolution replace the traditional convolution in Darknet-53, which can reduce the loss of features, decrease model parameters and increase the detection speed. Thirdly, multi-scale feature fusion structure is used to get more shallow information, so as to improve the accuracy of helmet wearing detection. Experimental results show that compared with YOLOv3, YOLOv3-He1met increases by 52% in FPS (Frame Detection Per Second) and 5.7% in mAP (Mean Average Precision). The algorithm improves the detection accuracy and speed of helmet wearing detection, which has certain practicality for production.