Helmet Wearing Detection Algorithm Based on Lightweight Network Model
Haikun Chang, Ying Hu · 2022
Aiming at the problems that the traditional helmet wearing detection algorithm has a large amount of parameters and calculation as well as poor real-time performance, which is not conducive to field deployment, helmet wearing detection algorithm called YOLOv4-MAC based on lightweight network model is proposed. Firstly, the YOLOv4 backbone feature extraction network CSPDarkNet53 is replaced by MobileNet V3, and the Depthwise Separable convolution(DW) is used to replace the traditional convolution in PANet and prediction network, which greatly reduces the amount of parameters and calculation; Then SPPNet is replaced by ASPPNet to increase the network receptive field and reduce the amount of network calculation; Finally, CBAM module is added before YOLOHead module to increase the detection ability of small targets. The experimental results show that the mAP of the improved algorithm is 87.82%, the detection speed is greatly improved, and the helmet wearing detection condition can be judged quickly and accurately.