An Improved Helmet Detection Algorithm Based on YOLO V4

Abstract

The existing helmet detection algorithms have disadvantages such as difficulty in detecting occluded targets, small targets, etc. To address those problems, a YOLO V4-based helmet detection improvement algorithm has been proposed. Firstly, the model’s backbone structure is improved, and the backbone’s multi-scale feature extraction capability is enhanced by using MCM modules with different sized convolutional kernels, the FSM channel attention module is used to guide the model to dynamically focus on the channel features of extracted small targets and obscured target information. Secondly, in order to optimize the model training, the latest loss function Eiou is used to replace Ciou for anchor frame regression prediction to improve the convergence speed and regression accuracy of the model. Finally, a helmet dataset is constructed from this paper, and a K-means clustering algorithm is used to cluster the helmet dataset and select the appropriate a priori candidate frames. The experimental results show that the improved algorithm has a significant improvement in detection accuracy compared with the original YOLO V4 algorithm, and can have a positive detection effect on small targets and obscured targets.