Research Article
Higher Accuracy Yolov5 Based Safety Helmet Detection
@INPROCEEDINGS{10.1007/978-3-031-36011-4_8, author={Zizhen Wang and Yuegong Sun and Zhening Wang and Ao Li}, title={Higher Accuracy Yolov5 Based Safety Helmet Detection}, proceedings={6GN for Future Wireless Networks. 5th EAI International Conference, 6GN 2022, Harbin, China, December 17-18, 2022, Proceedings, Part I}, proceedings_a={6GN}, year={2023}, month={7}, keywords={Deep Learning Object Detection YOLOv5}, doi={10.1007/978-3-031-36011-4_8} }- Zizhen Wang
Yuegong Sun
Zhening Wang
Ao Li
Year: 2023
Higher Accuracy Yolov5 Based Safety Helmet Detection
6GN
Springer
DOI: 10.1007/978-3-031-36011-4_8
Abstract
For the construction site with high-risk possibility, object detection based on safety helmet and reflective clothing will greatly reduce the risk of workers. At present, the algorithm based on deep learning is the mainstream algorithm of object detection. Among them, the YOLO algorithm is fast and widely used in real-time safety helmet detection. However, for the problems of small objects such as safety helmets and relatively dense detection scene objects, the detection effect is not ideal. For these problems, this paper proposes an improvement of the safety helmet detection algorithm based on YOLOv5s. The DenseBlock module is used in the improved algorithm to replace the Focus structure in the backbone network, which has an improved feature extraction capability for the network; secondly, Soft-NMS is used to retain more category frames when removing redundant frames. After the experiments, it is shown that the accuracy is improved on the homemade safety helmet dataset, which indicates the effectiveness of the improved algorithm.
