SUN Guangling,ZHOU Yunlong.Traffic Vehicle Detection based on Self-Attention Combined with Context Decoupling[J].Journal of Chengdu University of Information Technology,2024,39(04):422-429.[doi:10.16836/j.cnki.jcuit.2024.04.005]
自注意力结合上下文解耦的交通车辆检测
- Title:
- Traffic Vehicle Detection based on Self-Attention Combined with Context Decoupling
- 文章编号:
- 2096-1618(2024)04-0422-08
- 分类号:
- TP391.4
- 文献标志码:
- A
- 摘要:
- 为应对车流量、时间地点和天气等因素给交通车辆检测带来的挑战,提出基于YOLOv5s模型的新算法。该改进模型适用于各种交通场景,其改进如下:在特征串联阶段引入高效的二维局部特征叠加自注意力(ELFSa),以增强模型对目标的感知能力; 将YOLOv5s的检测头替换为简易特定于任务的上下文解耦(S-TSCODE),以实现分类和定位子任务之间的完美平衡,从而改善模型的收敛; 为减少运算负担,模型中的大于3×3的部分卷积操作被替换成了GSConv。实验结果显示,改进的YOLOv5s在各方面均有提升,其中mAP@0.5为98.9%,mAP@0.5:0.95为87.0%,分别提升0.1%和1.5%。针对各种复杂的交通场景,所提出的方法增强了车辆检测的性能和鲁棒性。
- Abstract:
- To address the challenges arising from factors like traffic flow, time, place, and weather on traffic vehicle detection, A novel algorithm, which builds upon the YOLOv5s model, has been introduced. This enhanced model demonstrates adaptability across diverse scenarios of transportation. The improvements are as follows: the incorporation of an efficient 2D local feature superimposed self-attention(ELFSa)during the feature series stage, aiming to enrich the model’s object perception capabilities; replace the detection head of YOLOv5s with a simple task-specific context decoupling(S-TSCODE)to achieve a perfect balance between classification and localization subtasks Balance, thereby improving the convergence of the model; To mitigate computational complexity, certain convolutions within the model with dimensions of 3×3 or larger are substituted with GSConv. The experimental findings demonstrate that the enhanced YOLOv5s model exhibits improvements across all aspects, particularly in terms of mAP@0.5 is 98.9%, and mAP@0.5:0.95 is 87.0%, which are respectively increased by 0.1% and 1.5%.Addressing a wide range of intricate traffic scenarios, the suggested methodology enhanced the performance and robustness of vehicle detection.
参考文献/References:
[1] 陆化普,李瑞敏.城市智能交通系统的发展现状与趋势[J].工程研究-跨学科视野中的工程,2014,6(1):6-19.
[2] 张富凯,杨峰,李策.基于改进YOLOv3的快速车辆检测方法[J].计算机工程与应用,2019,55(2):12-20.
[3] Cheng H Y,Weng C C,Chen Y Y.Vehicle detection in aerial surveillance using dynamic Bayesian networks[J].IEEE transactions on image processing,2011,21(4):2152-2159.
[4] Zhang J,Guo X,Zhang C,et al.A vehicle detection and shadow elimination method based on greyscale information,edge information,and prior knowledge[J].Computers & Electrical Engineering,2021,94:107366.
[5] Girshick R.Fast r-cnn[C].Proceedings of the IEEE international conference on computer vision.2015:1440-1448.
[6] Cai Z,Vasconcelos N.Cascade r-cnn: Delving into high quality object detection[C].Proceedings of the IEEE conference on computer vision and pattern recognition.2018:6154-6162.
[7] Dosovitskiy A,Beyer L,Kolesnikov A,et al.An image is worth 16x16 words:Transformers for image recognition at scale[J].arXiv preprint arXiv:2010.11929,2020.
[8] Li W,Huang L.YOLOSA:Object detection based on 2D local feature superimposed self-attention[J].arXiv preprint arXiv:2206.11825,2022.
[9] Zhuang J,Qin Z,Yu H,et al.Task-Specific Context Decoupling for ObjectDetection[J].arXiv preprint arXiv:2303.01047,2023.
[10] 毕鹏程,罗健欣,陈卫卫.轻量化卷积神经网络技术研究[J].计算机工程与应用,2019,55(16):25-35.
[11] Chollet F.Xception: Deep learning with depthwise separable convolutions[C].Proceedings of the IEEE conference on computer vision and pattern recognition.2017:1251-1258.
[12] Li H,Li J,Wei H,et al.Slim-neck by GSConv: A better design paradigm of detector architectures for autonomous vehicles[J].arXiv preprint arXiv:2206.02424,2022.
[13] Krizhevsky A,Sutskever I,Hinton G E.Imagenet classification with deep convolutional neural networks[J].Advances in neural information processing systems,2012,25.
[14] Liu S,Qi L,Qin H,et al.Path aggregation network for instance segmentation[C].Proceedings of the IEEE conference on computer vision and pattern recognition.2018:8759-8768.
[15] Hu J,Shen L,Sun G.Squeeze-and-excitation networks[C].Proceedings of the IEEE conference on computer vision and pattern recognition.2018:7132-7141.
[16] Woo S,Park J,Lee J Y,et al.Cbam:Convolutional block attention module[C].Proceedings of the European conference on computer vision(ECCV).2018:3-19.
[17] Ge Z,Liu S,Wang F,et al.Yolox:Exceeding yolo series in 2021[J].arXiv preprint arXiv:2107.08430,2021.
[18] Everingham M,Gool L V,Williams C K I,et al.The Pascal Visual Object Classes(VOC)Challenge[J].International Journal of Computer Vision,2010,88(2):303-338.
[19] Wen L,Du D,Cai Z,et al.UA-DETRAC: A new benchmark and protocol for multi-object detection and tracking[J].ComputerVision and Image Understanding,2020,193:102907.
[20] Liu S,Huang D,Wang Y.Learning spatial fusion for single-shot object detection[J].arXiv preprint arXiv:1911.09516,2019.
相似文献/References:
[1]李 静,鲜 林,王海江.基于YOLOv3的船只检测算法研究[J].成都信息工程大学学报,2023,38(01):37.[doi:10.16836/j.cnki.jcuit.2023.01.006]
LI Jing,XIAN Lin,WANG Haijiang.Research on Ship Detection Algorithm based on YOLOv3[J].Journal of Chengdu University of Information Technology,2023,38(04):37.[doi:10.16836/j.cnki.jcuit.2023.01.006]
[2]魏春梅,马尚昌,卢会国,等.基于视频识别的气象观测场设备监控技术研究[J].成都信息工程大学学报,2023,38(02):129.[doi:10.16836/j.cnki.jcuit.2023.02.001]
WEI Chunmei,MA Shangchang,LU Huiguo,et al.Research on Equipment Monitoring Technology of Meteorological Observation Field based on Video Recognition[J].Journal of Chengdu University of Information Technology,2023,38(04):129.[doi:10.16836/j.cnki.jcuit.2023.02.001]
备注/Memo
收稿日期:2023-08-15
基金项目:国家自然科学基金资助项目(62001004); 安徽省高校协同创新项目(GXXT-2021-024); 2023年安徽省住房城乡建设科学技术计划资助项目(2023-YF058、2023-YF113)
通信作者:孙光灵.E-mail:sunguangling@163.com