基于YOLOv5改进的交通场景目标检测算法HCAM-YOLO

doi:10.13474/j.cnki.11-2246.2024.1111

摘要/Abstract

摘要： 交通场景下快速、准确的目标检测对智慧交通管理、智能驾驶路径决策具有重要意义。由于交通环境的复杂多变性及目标特征的多样性和稀疏性,传统的目标检测模型往往面临检测精度不足、漏检与误检率高的问题。为了克服这些困难,本文提出一种基于HcPAN特征融合网络的YOLOv5目标检测模型HCAM-YOLO。该方法的核心是针对PAN网络特征融合过程中局部信息易丢失的问题,设计了一种混合卷积注意力机制(HCAM),用于特征融合网络中的多尺度信息增强提取。将HCAM模块嵌入PAN底层结构,以提升局部关键特征敏感性,强化深层语义信息与浅层位置信息的融合效果。该方法的创新性在于利用注意力机制优化特征融合过程,从而提升模型在复杂交通环境中对行人、机动车辆等目标的检测性能。试验数据集由Rope 3D数据集与自制车流数据集Road Veh、自制行人数据集Road Ped联合组成。试验结果表明,HCAM模块比其他注意力机制更适合嵌入PAN底层网络中,最终算法模型HCAM-YOLO相比于基础模型YOLOv5,精确率和召回率分别提高3.4%和3.2%;平均精度均值提高3.8%。本文算法模型对复杂背景下交通场景目标检测任务具有良好的适应性。

关键词: 目标检测, 交通场景, 特征增强, 注意力机制, YOLOv5

Abstract: The rapid and precise detection of targets in traffic scenarios is crucial for intelligent traffic management and driving path decision-making. Traditional target detection models often grapple with issues such as inadequate detection accuracy, high rates of leak detection and false detection due to the complexity and variability of the traffic environment, and the diversity and sparsity of target features. To address these challenges, this paper introduces a YOLOv5 target detection model, HCAM-YOLO, which leverages the HcPAN feature fusion network. The crux of this approach lies in addressing the issue of local information being easily lost during the PAN network's feature fusion process. A hybrid convolutional attention mechanism(HCAM) is designed to enhance multi-scale information extraction in feature fusion networks. By integrating the HCAM module into the PAN's underlying structure, the sensitivity of key local features is enhanced, while the fusion effect of deep semantic information and shallow positional data is strengthened. This method's novelty lies in its use of an attention mechanism to optimize the feature fusion process, thereby improving the model's detection performance of pedestrians, motor vehicles, and other targets in complex traffic environments. The experimental dataset comprises the Rope 3D dataset, Road Veh dataset, and Road Ped dataset. The results demonstrate that the HCAM module is more suitable for integration into the underlying PAN network than other attention mechanisms. When compared to the basic YOLOv5 model, the precision and recall of the final HCAM-YOLO algorithm model increased by 3.4% and 3.2%,respectively, and mAP@0.5/% by 3.8%. The HCAM-YOLO algorithm model proposed in this paper exhibits strong adaptability to target detection tasks in traffic scenes with complex backgrounds.

Key words: object detection, traffic scene, feature enhancement, attention mechanism, YOLOv5

中图分类号:

P208

王志涛, 张瑞菊, 王坚, 赵佳星, 刘严涛. 基于YOLOv5改进的交通场景目标检测算法HCAM-YOLO[J]. 测绘通报, 2024, 0(11): 61-67.

WANG Zhitao, ZHANG Ruiju, WANG Jian, ZHAO Jiaxing, LIU Yantao. Improved object detection algorithm HCAM-YOLO in traffic scenes based on YOLOv5[J]. Bulletin of Surveying and Mapping, 2024, 0(11): 61-67.

参考文献

[1] ADOUANI A, BEN HENIA W M, LACHIRI Z.A comparison of face detection methods using spontaneous videos[J].Multimedia Tools and Applications, 2022, 81(16):23163-23191.
[2] SUNKARA R, LUO Tie.No more strided convolutions or Pooling:a new CNN building block for Low-resolution images and Small objects[M]//Machine Learning and Knowledge Discovery in Databases.Cham:Springer Nature Switzerland, 2023:443-459.
[3] TANG Lindong, YUN Lijun, CHEN Zaiqing, et al.HRYNet:a highly robust YOLO network for complex road traffic object detection[J].Sensors, 2024, 24(2):642.
[4] ZHANG Yuan, SUN Youpeng, WANG Zheng, et al.YOLOv7-RAR for urban vehicle detection[J].Sensors, 2023, 23(4):1801.
[5] SONG Yingkun, HONG Shunhe, HU Chentao, et al.MEB-YOLO:an efficient vehicle detection method in complex traffic road scenes[J].Computers, Materials & Continua, 2023, 75(3):5761-5784.
[6] WANG Junfan, CHEN Yi, DONG Zhekang, et al.Improved YOLOv5 network for real-time multi-scale traffic sign detection[J].Neural Computing and Applications, 2023, 35(10):7853-7865.
[7] ZHU Linlin, GENG Xun, LI Zheng, et al.Improving YOLOv5 with attention mechanism for detecting boulders from planetary images[J].Remote Sensing, 2021, 13(18):3776.
[8] DONG Hao, YUAN Mu, WANG Shu, et al.PHAM-YOLO:a parallel hybrid attention mechanism network for defect detection of meter in substation[J].Sensors, 2023, 23(13):6052.
[9] GUO Z, WANG C, YANG G, et al.MSFT-YOLO:improved YOLOv5 based on transformer for detecting defects of steel surface[J].Sensors, 2022, 22(9):3467.
[10] LI Shaoxiong, KONG Fanning, WANG Ruoqi, et al.EFD-YOLOv4:a steel surface defect detection network with encoder-decoder residual block and feature alignment module[J].Measurement, 2023, 220:113359.
[11] QIAN Xiaohong, WANG Xu, YANG Shengying, et al.LFF-YOLO:a YOLO algorithm with lightweight feature fusion network for multi-scale defect detection[J].IEEE Access, 2022, 10:130339-130349.
[12] WANG Yang, WANG Hongyuan, XIN Zihao.Efficient detection model of steel strip surface defects based on YOLO-V7[J].IEEE Access, 2022, 10:133936-133944.
[13] ZHAO Chao, SHU Xin, YAN Xi, et al.RDD-YOLO:a modified YOLO for detection of steel surface defects[J].Measurement, 2023, 214:112776.
[14] YE Xiang, HE Zihang, HENG Wang, et al.Toward understanding the effectiveness of attention mechanism[J].AIP Advances, 2023, 13(3):39-47.
[15] HOWARD A, SANDLER M, CHEN Bo, et al.Searching for MobileNetV3[C]//Proceedings of 2019 IEEE/CVF International Conference on Computer Vision.Seoul, Korea (South):IEEE, 2019.