基于改进DeepSORT算法的摩托车头盔佩戴检测

doi:10.3778/j.issn.1002-8331.2205-0163

摘要/Abstract

摘要： 为了解决在实际道路场景中对摩托车驾乘者头盔佩戴情况的检测与跟踪问题，提出一种优化的DeepSORT多目标跟踪摩托车头盔检测算法。使用改进的YOLOv3算法作为目标检测器，该方法采用MobileNetv3-Large作为YOLOv3特征提取网络，以轻量化网络结构，借鉴CEM模块的思想，引入SoftPool池化层和SE模块，构建了深浅语义融合增强模块，加强网络对输入特征深浅语义的编码能力。为了实现摩托车头盔佩戴分类和DeepSORT算法中外观特征识别模型的优化，构建多任务学习框架，通过对比损失函数和改进的交叉损失函数对多任务网络进行迭代训练来提升相似度学习能力和实现最终头盔佩戴情况分类。实验结果表明：在目标检测方面，改进网络的摩托车检测精度相较原始算法提升了4.56个百分点，模型大小仅为YOLOv3的13.7%。结合DeepSORT算法，多目标跟踪准确率相较于YOLOv3-DeepSORT算法从51.6%提升到了67.5%，多目标跟踪精度从57.3%提升到75.6%，检测追踪速度由6?FPS提升到了20?FPS，最终的NPH模型检测分类精度为72.42%。实现了对实际道路中摩托车驾乘人员头盔佩戴检测及追踪，具有较强的实用价值。

关键词: YOLOv3, DeepSORT, 多任务学习, 摩托车头盔佩戴检测

Abstract: To solve the problem of detecting and tracking the helmet-wearing of motorcycle riders in the actual road scene, an optimized DeepSORT multi-target tracking motorcycle helmet detection algorithm is proposed. Using the improved YOLOv3 algorithms for the target detector, this method first uses MobileNetv3-Large as the YOLOv3 feature extraction network to lighten the network structure, and then draws on the idea of the CEM module, introduces the SoftPool pooling layer and SE. The deep and shallow semantic fusion enhancement module is constructed to enhance the network’s ability to encode the deep and shallow semantics of the input features. To realize the optimization of the motorcycle helmet wearing classification and the appearance feature recognition model in the DeepSORT algorithm, a multi-task learning framework is constructed. The multi-task network is iteratively trained by comparing the loss function and the improved cross loss function to improve the similarity learning ability and achieve the final helmet wearing situation classification. The experimental results show that: in terms of target detection, the motorcycle detection accuracy of the improved network is improved by 4.56?percentage points compared with the original algorithm, and the model size is only 13.7% of that of YOLOv3. Combined with the DeepSORT algorithm, the multi-target tracking accuracy has increased from 51.6% to 67.5% compared with the YOLOv3-DeepSORT algorithm, the multi-target tracking accuracy has increased from 57.3% to 75.6%, and the detection and tracking speed has increased from 6?FPS to 20?FPS. The final NPH model detection classification accuracy is 72.42%. It realizes the detection and tracking of helmet wearing of motorcycle drivers on actual roads and has strong practical value.

Key words: YOLOv3, DeepSORT, multi-task learning, motorcycle helmet wearing detection

冉险生, 张之云, 陈卓, 苏山杰, 陈俊豪. 基于改进DeepSORT算法的摩托车头盔佩戴检测[J]. 计算机工程与应用, 2023, 59(5): 194-204.

RAN Xiansheng, ZHANG Zhiyun, CHEN Zhuo, SU Shanjie, CHEN Junhao. Motorcycle Helmet Wearing Detection Based on Improved DeepSORT Algorithm[J]. Computer Engineering and Applications, 2023, 59(5): 194-204.

参考文献

[1] 余松凌.摩托车头盔佩戴行为研究[D].成都：西南交通大学，2018.
YU S L.Research on motorcycle helmet wearing behavior[D].Chengdu：Southwest Jiaotong University，2018.
[2] 张瑶，卢焕章，张路平，等.基于深度学习的视觉多目标跟踪算法综述[J].计算机工程与应用，2021，57（13）：55-66.
ZHANG Y，LU H Z，ZHANG L P，et al.A review of visual multi-target tracking algorithms based on deep learning[J].Computer Engineering and Applications，2021，57（13）：55-66.
[3] ZOU Z，SHI Z，GUO Y，et al.Object detection in 20 years：a survey[J].arXiv：1905.05055，2019.
[4] 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.
[5] GIRSHICK R.Fast R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision，2015：1440-1448.
[6] REN S，HE K，GIRSHICK R，et al.Faster R-CNN：towards real-time object detection with region proposal networks[C]//Advances in Neural Information Processing Systems，2015：91-99.
[7] LIU W，ANGUELOV D，ERHAN D，et al.SSD：single shot multibox detector[C]//European Conference on Computer Vision，2016：21-37.
[8] REDMON J，FARHADI A.Yolov3：an incremental improvement[J].arXiv：1804.02767，2018.
[9] BOCHKOVSKIY A，WANG C Y，LIAO H Y M.Yolov4：optimal speed and accuracy of object detection[J].arXiv：2004.10934，2020.
[10] Ultralytics.YOLOv5[EB/OL].[2021].https//github.com/ultralytics/yolov5.
[11] ESPINOSA J E，VELASTIN S A，BRANCH J W.Motorcycle detection and classification in urban scenarios using a model based on Faster R-CNN[J].arXiv：1808.02299，2018.
[12] 刘琛，王江涛，王明阳.引入视觉机制的SSD网络在摩托车头盔佩戴检测中的应用[J].电子测量与仪器学报，2021，35（3）：144-151.
LIU C，WANG J T，WANG M Y.Application of SSD network introducing visual mechanism in motorcycle helmet wearing detection[J].Journal of Electronic Measurement and Instrumentation，2021，35（3）：144-151.
[13] BOONSIRISUMPUN N，PUARUNGROJ W，WAIROT-CHANAPHUTTHA P.Automatic detector for bikers with no helmet using deep learning[C]//IEEE International Computer Science and Engineering Conference（ICSEC），2018.
[14] MAHMOUDI N，AHADI S M，RAHMATI M.Multi-target tracking using CNN-based features：CNNMTT[J].Multimedia Tools and Applications，2019，78（6）：7077-7096.
[15] HUO W，OU J，LI T.Multi-target tracking algorithm based on deep learning[C]//Journal of Physics：Conference Series，2021，1948（1）：012011.
[16] WOJKE N，BEWLEY A，PAULUS D.Simple online and realtime tracking with a deep association metric[C]//2017 IEEE International Conference on Image Processing（ICIP），2017：3645-3649.
[17] BEWLEY A，GE Z，OTT L，et al.Simple online and realtime tracking[C]//2016 IEEE International Conference on Image Processing（ICIP），2016：3464-3468．
[18] QIN Z，ZE M L，ZHAO N，et al.ThunderNet：towards real-time generic object detection on mobile devices[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision，2019：6718-6727.
[19] HE K，ZHANG X，REN S，et al.Spatial pyramid pooling in deep convolutional networks for visual recognition[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2015，37（9）：1904-1916.
[20] STERGIOU A，POPPE R，KALLIATAKIS G.Refining activation downsampling with SoftPool[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision，2021：10357-10366.
[21] CARUANA R.Multitask learning：a knowledge-based source of inductive bias[C]//Proceedings of the Tenth International Conference on Machine Learning，1993：41-48.
[22] BROMLEY J，GUYON I，LECUN Y，et al.Signature verification using a “siamese” time delay neural network[C]//Advances in Neural Information Processing Systems，1993.
[23] ZHOU D，HOU Q，CHEN Y，et al.Rethinking bottleneck structure for efficient mobile network design[J].arXiv：2007.02269，2020.