Safety Helmet Wearing Detection Based on Improved YOLOv5s

Abstract

For construction sites in high-risk industries such as the construction industry, wearing a helmet can minimize head injuries. Aiming at the low detection accuracy of the existing detection algorithms for wearing helmets, and the detection of small objects in complex and dense scenes is prone to false detection and missed detection, an improved helmet detection algorithm based on YOLOv5 is proposed. First, the anchors are re-clustered by the K-means++ algorithm to obtain an anchor box size that is more suitable for the helmet dataset. Secondly, the Coordinate Attention module is integrated into the backbone network of YOLOv5, so that the network can obtain more detailed features, thus, the feature extraction capability of the network is enhanced. Finally, the CIOU loss function of the bounding box regression is replaced with the EIOU loss function to realize the fusion of more bounding box information, thereby improving the accuracy of the algorithm prediction. The experimental results show that the mAP of the improved YOLOv5 reaches 95.0%, which is 3.2 percentage points higher than the original model. The improved YOLOv5 algorithm meets the accuracy requirements of detecting small objects in more complex and dense scenes, and can be more accurate. It can be better applied to practical scenarios.