Helmet-YOLO：一种更高精度的道路安全头盔检测算法

doi:10.3778/j.issn.1002-8331.2406-0013

摘要/Abstract

摘要： 针对现有的道路安全头盔检测算法受背景环境影响较大，面对遮挡以及目标与环境相似等检测场景检测精度不高的问题，从特征融合和损失计算的角度，开发了一种新的Helmet-YOLO架构。利用渐进式特征金字塔网络结构降低多尺度特征融合过程中存在的巨大语义差距，提升算法在复杂场景下的检测能力。同时，提出的PCAHead检测头和HelmetIoU边界框损失函数优化了模型理解和处理数据的能力，提高了模型损失计算的效率和精度，加速了模型的收敛。实验结果表明，Helmet-YOLOn算法和Helmet-YOLOs算法的mAP@50分别提升了3.7和2.9个百分点，优于实验中的所有同尺度模型，另外Helmet-YOLO的大尺度模型在延迟和精度上也优于多数实验模型。实验证明Helmet-YOLO算法具有更高的精度和鲁棒性，更适合复杂场景的道路安全头盔检测。

关键词: 头盔检测, Helmet-YOLO, 渐进式特征金字塔网络, PCAHead检测头, HelmetIoU

Abstract: In view of the significant influence of background environment on the existing road safety helmet detection algorithm and the issue of low detection accuracy in detection scenarios such as occlusion and similarity between targets and the environment, this paper considers the YOLOv8 model from the perspectives of feature fusion and loss calculation. By utilizing the progressive feature pyramid network structure to reduce the substantial semantic gap in the process of multi-scale feature fusion, the algorithm’s detection capability in complex scenes is enhanced. Additionally, the proposed PCAHead for detection and HelmetIoU bounding box loss function optimize the model’s understanding and data processing capabilities, improving the efficiency and accuracy of model loss calculation, thereby accelerating model convergence. Experimental results show that the mAP@50 of the Helmet-YOLOn algorithm and Helmet-YOLOs algorithm have increased by 3.7 percentage points and 2.9 percentage points, respectively, outperforming all models of the same scale in the experiment. Furthermore, the large-scale model of Helmet-YOLO also outperforms most experimental models in terms of latency and accuracy. The experiments demonstrate that the Helmet-YOLO algorithm has higher accuracy and robustness, making it more suitable for road safety helmet detection in complex scenarios.

Key words: helmet detection, Helmet-YOLO, asymptotic feature pyramid network, PCAHead detection head, HelmetIoU

周顺勇, 彭梓洋, 张航领, 胡琴, 陆欢, 张宗良. Helmet-YOLO：一种更高精度的道路安全头盔检测算法[J]. 计算机工程与应用, 2025, 61(2): 135-144.

ZHOU Shunyong, PENG Ziyang, ZHANG Hangling, HU Qin, LU Huan, ZHANG Zongliang. Helmet-YOLO: Higher-Accuracy Road Safety Helmet Detection Algorithm[J]. Computer Engineering and Applications, 2025, 61(2): 135-144.

参考文献

[1] ORGANIZATION W H. Global status report on road safety 2018[Z]. World Health Organization, 2019.
[2] WARANUSAST R, BUNDON N, TIMTONG V, et al. Machine vision techniques for motorcycle safety helmet detection[C]//Proceedings of the 28th International Conference on Image and Vision Computing, 2013: 35-40.
[3] SILVA R, AIRES K, SANTOS T, et al. Automatic detection of motorcyclists without helmet[C]//Proceedings of the 2013 XXXIX Latin American Computing Conference, 2013: 1-7.
[4] 张阳婷, 黄德启, 王东伟, 等. 基于深度学习的目标检测算法研究与应用综述[J]. 计算机工程与应用, 2023, 59(18): 1-13.
ZHANG Y T, HUANG D Q, WANG D W, et al. Review on research and application of deep learning-based target detection algorithms[J]. Computer Engineering and Applications, 2023, 59(18): 1-13.
[5] GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2014: 580-587.
[6] GIRSHICK R. Fast R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision, 2015: 1440-1448.
[7] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 39(6): 1137-1149.
[8] LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]//Proceedings of the 14th European Conference on Computer Vision, 2016: 21-37.
[9] 谢溥轩, 崔金荣, 赵敏. 基于改进YOLOv5的电动车头盔佩戴检测算法[J]. 计算机科学, 2023, 50(1): 420-425.
XIE P X, CUI J R, ZHAO M. Electiric bike helment wearing detection alogrithm based on improved YOLOv5[J]. Computer Science, 2023, 50(1): 420-425.
[10] 张鑫, 周顺勇, 李思诚, 等. 基于注意力机制和跨尺度特征融合的摩托车头盔检测算法[J]. 电子测量技术, 2023, 46(12): 134-142.
ZHANG X, ZHOU S Y, LI S C, et al. Motorcycle helmet detection algorithm based on attention mechanism and cross-scale feature fusion[J]. Electronic Measurement Technology, 2023, 46(12): 134-142.
[11] HUANG B, WU S, XIANG X, et al. An improved YOLOv5s-based helmet recognition method for electric bikes[J]. Applied Sciences, 2023, 13(15): 8759.
[12] SHABESTARI Z B, HOSSEININAVEH A, REMONDINO F. Motorcycle detection and collision warning using monocular images from a vehicle[J]. Remote Sensing, 2023, 15(23): 5548.
[13] CHEN S, LAN J, LIU H, et al. Helmet wearing detection of motorcycle drivers using deep learning network with residual transformer-spatial attention[J]. Drones, 2022, 6(12): 415.
[14] JIA W, XU S, LIANG Z, et al. Real‐time automatic helmet detection of motorcyclists in urban traffic using improved YOLOv5 detector[J]. IET Image Processing, 2021, 15(14): 3623-3637.
[15] SZEGEDY C, LIU W, JIA Y, et al. Going deeper with convolutions[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015: 1-9.
[16] XIE S, GIRSHICK R, DOLLáR P, et al. Aggregated residual transformations for deep neural networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 5987-5995.
[17] ENGELMAYER V, GEORGIEV D G, VELI?KOVI? P. Parallel algorithms align with neural execution[C]//Proceedings of the Learning on Graphs Conference, 2023: 31.
[18] WEI J, ZHANG X, JI Z, et al. Deploying and scaling distributed parallel deep neural networks on the Tianhe-3 prototype system[J]. Scientific Reports, 2021, 11(1): 20244.
[19] WU J, SUN Y. Evolving deep parallel neural networks for multi-task learning[C]//Proceedings of the International Conference on Algorithms and Architectures for Parallel Processing, 2021: 517-531.
[20] DAI Y, WANG J, LI J, et al. PDBNet: parallel dual branch network for real-time semantic segmentation[J]. International Journal of Control, Automation and Systems, 2022, 20(8): 2702-2711.
[21] WANG Y, ERGEN T, PILANCI M. Parallel deep neural networks have zero duality gap[J]. arXiv:2110.06482, 2021.
[22] LIU S, HUANG D, WANG Y. Learning spatial fusion for single-shot object detection[J]. arXiv:1911.09516, 2019.
[23] HE K, GKIOXARI G, DOLLáR P, et al. Mask R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision, 2017.
[24] REDMON J, FARHADI A. YOLO9000: better, faster, stronger[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 6517-6525.
[25] REDMON J, FARHADI A. YOLOv3: an incremental improvement[J]. arXiv:1804.02767, 2018.
[26] BOCHKOVSKIY A, WANG C, LIAO H M. YOLOv4: Optimal speed and accuracy of object detection[J]. arXiv:2004.
10934, 2020.
[27] WU W, LIU H, LI L, et al. Application of local fully convolutional neural network combined with YOLOv5 algorithm in small target detection of remote sensing image[J]. PloS one, 2021, 16(10): 259283.
[28] WANG C, BOCHKOVSKIY A, LIAO H M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023: 7464-7475.
[29] WANG C, YEH I, LIAO H M. YOLOv9: learning what you want to learn using programmable gradient information[J]. arXiv: 2402. 13616, 2024.
[30] ZHANG H, CHANG H, MA B, et al. Dynamic R-CNN: towards high quality object detection via dynamic training[C]//Proceedings of the 16th European Conference on Computer Vision, 2020: 260-275.
[31] CARION N, MASSA F, SYNNAEVE G, et al. End-to-end object detection with transformers[C]//Proceedings of the European Conference on Computer Vision, 2020: 213-229.
[32] ZHAO Y, LV W, XU S, et al. Detrs beat YOLOs on real-time object detection[J]. arXiv: 2304. 08069, 2023.