使用场景增强的安全帽佩戴检测方法研究

doi:10.3778/j.issn.1002-8331.2103-0030

摘要/Abstract

摘要： 为了解决现有安全帽佩戴数据集样本数量有限导致模型检测精度较低的问题，提出一种基于场景增强的样本扩充算法。该算法将训练集中随机抽取的图像中的检测目标随机缩放后，粘贴到另一随机场景图像上的任意位置，基于现有场景构建出拥有新的检测目标的增强场景，通过场景增强扩充安全帽佩戴训练数据集，增加训练数据集的多样性。为了验证该算法在安全帽佩戴检测中的有效性，使用场景增强算法扩充HelmetWear数据集，并用其训练基于YOLO v4的安全帽佩戴检测模型，通过检测精度评估场景增强算法。在HelmetWear数据集上检测精度达到93.81%，检测精度提升了6.39个百分点。实验结果表明该算法能有效提升安全帽佩戴检测的精度，特别是在样本最为欠缺的小目标上表现更为显著；场景增强算法对解决其他领域目标检测训练数据不足的问题有重要的借鉴意义。

关键词: 安全帽检测, 场景增强, HelmetWear数据集, YOLO v4

Abstract: In order to solve the problem of low model detection accuracy caused by the limited number of samples in the existing helmet dataset, a sample expansion algorithm based on scene augment is proposed. This algorithm randomly scales the small and medium targets in the foreground image randomly extracted from the training set, and pastes it to any position on another random scene image to construct a new real training image with border labels to improve the combination of background and foreground in the dataset diversity. In order to evaluate the effectiveness of the algorithm in the helmet wearing detection, the HelmetWear dataset is expanded using the scene augment algorithm and the scene augment algorithm is evaluated by detection accuracy using its training hard hat wearing detection model based on YOLO v4. Detection accuracy reaches 93.81% on the HelmetWear dataset and improves detection accuracy by 6.39 percentage points. The experimental results show that the algorithm can effectively improve the accuracy of hard hat wearing detection, especially on the small targets where the sample is most lacking. The scene augment algorithm is of great significance to solve the problem of insufficient target detection training data in other fields.

Key words: safety helmet detection, scene-augment, HelmetWear dataset, YOLO v4

徐传运, 袁含香, 李刚, 郑宇, 刘欢. 使用场景增强的安全帽佩戴检测方法研究[J]. 计算机工程与应用, 2022, 58(19): 326-332.

XU Chuanyun, YUAN Hanxiang, LI Gang, ZHENG Yu, LIU Huan. Research on Safety Helmet Wearing Detection Method Based on Scene Augment[J]. Computer Engineering and Applications, 2022, 58(19): 326-332.

参考文献

[1] 住房和城乡建设部办公厅关于2019年房屋市政工程生产安全事故情况的通报[EB/OL].（2020-06-19）[2021-02-20].http：//www.mohurd.gov.cn/wjfb/202006/t20200624_246031.html.
Circular of the General Office of the Ministry of Housing and Urban Rural Development on production safety accidents of municipal housing projects in 2019.[EB/OL].（2020-06-19）[2021-02-20].http：//www.mohurd.gov.cn/wjfb/202006/t20200624_246031.html.
[2] 言有三.深度学习中的数据增强方法都有哪些？[EB/OL].（2019-04-22）[2021-02-20].https：//blog.csdn.net/weixin_43876801/article/details/102958247.
YAN Y S.What are the data enhancement methods in deep learning?[EB/OL].（2019-04-22）[2021-02-20].https：//blog.csdn.net/weixin_43876801/article/details/102958247.
[3] DWIBEDI D，MISRA I，HEBERT M.Cut，paste and learn：Surprisingly easy synthesis for instance detection[C]//Proceedings of 2017 IEEE International Conference on Computer Vision（ICCV），2017.
[4] ZHANG H，CISSE M，DAUPHIN Y N，et al.Mixup：Beyond empirical risk minimization[J].arXiv：1710.09412，2017.
[5] INOUE H.Data augmentation by pairing samples for images classification[J].arXiv：1801.02929，2018.
[6] HUSSEIN A S，DIALLO B，LIU J.EBSMOTE：Evaluation-based synthetic minority oversampling Technique for imbalanced dataset learning[C]//Proceedings of IEEE 14th International Conference on Intelligent Systems and Knowledge Engineering（ISKE），2019：1226-1233.
[7] 张家伟，郭林明，杨晓梅.针对不平衡数据的过采样和随机森林改进算法[J].计算机工程与应用，2020，56（11）：39-45.
ZHANG J W，GUO L M，YANG X M.Improved oversampling and random forest algorithm for imbalanced data[J].Computer Engineering and Applications，2020，56（11）：39-45.
[8] BOCHKOVSKIY A，WANG C Y，LIAO H Y M.YOLOv4：Optimal speed and accuracy of object detection[J].arXiv：2004.10934，2020.
[9] 杜卉然，许亮，吕帅.基于邻域差分滤波生成式对抗网络的数据增强方法[J].计算机应用研究，2020，37（6）：301-305.
DU H R，XU L，LYU S.Data enhancement method based on neighborhood differential filter generative countermeasure network[J].Application Research of Computers，2020，37（6）：301-305.
[10] 孙晓，丁小龙.基于生成对抗网络的人脸表情数据增强方法[J].计算机工程与应用，2020，56（4）：115-121.
SUN X，DING X L.Data augmentation method based on generative adversarial networks for facial expression recognition sets[J].Computer Engineering and Applications，2020，56（4）：115-121.
[11] CUBUK E D，ZOPH B，MANé D，et al.AutoAugment：Learning augmentation strategies from data[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition，2019：113-123.
[12] 方明，孙腾腾，邵桢.基于改进YOLOv2的快速安全帽佩戴情况检测[J].光学精密工程，2019（5）：1196-1205.
FANG M，SUN T T，SHAO Z.Fast helmet wearing condition detection based on improved YOLOv2[J].Optics and Precision Engineering，2019（5）：1196-1205.
[13] LIN T，MAIRE M，BELONGIE S，et al.Microsoft COCO：Common objects in context[C]//Proceedings of European Conference on Computer Vision，2014：740-755.
[14] LOSHCHILOV I，HUTTER F.SGDR：Stochastic gradient descent with restarts，2016[J].arXiv：1608.03983，2016.
[15] ZEILER M D.ADADELTA：An adaptive learning rate method[J].arXiv：1212.5701，2012.
[16] MO X，WEI T，ZHANG H，et al.Label-smooth learning for fine-grained visual categorization[C]//Proceedings of Asian Conference on Pattern Recognition，2019：17-31.
[17] ZHONG Z，ZHENG L，KANG G，et al.Random erasing data augmentation[C]//Proceedings of the AAAI Conference on Artificial Intelligence，2017.
[18] REDMON J，FARHADI A.YOLO9000：Better，faster，stronger[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition，2017：6517-6525.
[19] REDMON J，FARHADI A.YOLOv3：An incremental improvement[J].arXiv：1804.02767，2018.