改进YOLOv8的轻量化密集行人检测方法

doi:10.3778/j.issn.1002-8331.2411-0437

摘要/Abstract

摘要： 针对密集行人检测存在小目标检测精度低、模型复杂的问题，提出一种改进YOLOv8的轻量化密集行人检测方法。引入DualConv模块替换原始Conv模块，帮助更深的卷积层更有效地提取信息，减少计算冗余并提高检测精度；通过融合RepViTBlock结构和分离与增强注意力机制SEMA（separated and enhancement attention）改进C2f，构建RS-C2f结构，提升模型的泛化和特征融合能力，并降低参数量；设计全新的空间金字塔模块SPPELAN_BiFPN，使模型对小目标行人检测精度显著提高，同时优化计算效率；采用Focal_Shape-IoU作为边界框回归损失函数，加快网络的收敛速度，提高对小目标的检测准确率。实验结果表明，改进模型的mAP@0.5、Precision和Recall在CrowdHuman数据集上提升2.4、1.1和2.1个百分点，在WiderPerson数据集上提升1.3、1.0和1.7个百分点，同时参数量下降39.6%。在嵌入式设备上单帧图像平均运行时间为55.1 ms，平均精度为90.7%，召回率为82.9%，表明改进模型在保证轻量化的同时提升了检测精度和速度。

关键词: 密集行人检测, 轻量化, RS-C2f, SPPELAN_BiFPN模块, Focal_Shape-IoU

Abstract: To address the challenges of low small object detection accuracy and model complexity in dense pedestrian detection, a lightweight improved YOLOv8 method for dense pedestrian detection is proposed. The DualConv module is introduced to replace the original Conv module, helping deeper convolutional layers more effectively extract information, reducing computational redundancy, and improving detection accuracy. By integrating the RepViTBlock structure and the separated and enhancement attention (SEMA) mechanism to improve C2f, the RS-C2f structure is constructed to enhance the model’s generalization and feature fusion capabilities while reducing the parameter count. A new spatial pyramid module, SPPELAN_BiFPN, is designed to significantly improve the model’s small object pedestrian detection accuracy while optimizing computational efficiency. Focal_Shape-IoU is employed as the bounding box regression loss function to accelerate network convergence and improve small object detection accuracy. Experimental results show that the improved model achieves a 2.4, 1.1, and 2.1 percentage points improvement in mAP@0.5, Precision, and Recall on the CrowdHuman dataset, respectively, and a 1.3, 1.0, and 1.7 percentage points improvement on the WiderPerson dataset, while the parameter count decreases by 39.6%. On embedded devices, the average inference time per image is 55.1 ms, with an average accuracy of 90.7% and recall rate of 82.9%, demonstrating that the improved model enhances detection accuracy and speed while maintaining lightweight characteristics.

Key words: dense pedestrian detection, lightweighting, RS-C2f, SPPELAN_BiFPN module, Focal_Shape-IoU

姚聪, 方遒, 郭星浩. 改进YOLOv8的轻量化密集行人检测方法[J]. 计算机工程与应用, 2025, 61(13): 138-150.

YAO Cong, FANG Qiu, GUO Xinghao. Improved Lightweight Dense Pedestrian Detection Method Based on YOLOv8[J]. Computer Engineering and Applications, 2025, 61(13): 138-150.

参考文献

[1] 张英俊, 白小辉, 谢斌红. CNN-Transformer特征融合多目标跟踪算法[J]. 计算机工程与应用, 2024, 60(2): 180-190.
ZHANG Y J, BAI X H, XIE B H. Multi-object tracking algorithm based on CNN-Transformer feature fusion[J]. Computer Engineering and Applications, 2024, 60(2): 180-190.
[2] 潘泽志.不同光线环境下的车辆检测及灯语信息识别研究[D]. 深圳: 深圳大学, 2022.
PAN Z Z. Vehicle detection and lamp signal recognition in different light environments[D]. Shenzhen: Shenzhen University, 2022.
[3] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[C]//Proceedings of the 25th International Conference on Neural Information Processing Systems, 2012: 1097-1105.
[4] GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2014: 580-587.
[5] GIRSHICK R. Fast R-CNN[C]//Proceedings of the 2015 IEEE International Conference on Computer Vision. Piscataway: IEEE, 2015: 1440-1448.
[6] REN S Q, HE K M, 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, 2017, 39(6): 1137-1149.
[7] LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]//Proceedings of the European Conference on Computer Vision. Cham: Springer, 2016: 21-37.
[8] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2016: 779-788.
[9] 韩文静, 何宁, 刘圣杰, 等. 基于改进ResNet-CrowdDet的密集行人检测算法[J]. 计算机工程与应用, 2023, 59(16): 196-204.
HAN W J, HE N, LIU S J, et al. Dense pedestrian detection algorithm based on improved ResNet-CrowdDet[J]. Computer Engineering and Applications, 2023, 59(16): 196-204.
[10] 贺宇哲, 徐光美, 何宁, 等. 迭代Faster R-CNN的密集行人检测[J]. 计算机工程与应用, 2023, 59(21): 214-221.
HE Y Z, XU G M, HE N, et al. Dense pedestrian detection with iterative faster R-CNN[J]. Computer Engineering and Applications, 2023, 59(21): 214-221.
[11] 高强, 唐福兴, 李栋, 等. 基于改进YOLOv5的密集场景行人检测方法研究[J]. 国外电子测量技术, 2023, 42(4): 125-130.
GAO Q, TANG F X, LI D, et al. Research on pedestrian detection method in dense scene based on improved YOLOv5[J]. Foreign Electronic Measurement Technology, 2023, 42(4): 125-130.
[12] 于范, 张菁. 滑窗注意力多尺度均衡的密集行人检测算法[J]. 计算机科学与探索, 2024, 18(5): 1286-1300.
YU F, ZHANG J. Dense pedestrian detection based on shifted window attention multi-scale equalization[J]. Journal of Frontiers of Computer Science and Technology, 2024, 18(5): 1286-1300.
[13] 高昂, 梁兴柱, 夏晨星, 等. 一种改进YOLOv8的密集行人检测算法[J]. 图学学报, 2023, 44(5): 890-898.
GAO A, LIANG X Z, XIA C X, et al. A dense pedestrian detection algorithm with improved YOLOv8[J]. Journal of Graphics, 2023, 44(5): 890-898.
[14] 胡倩,皮建勇,胡伟超,等.基于改进YOLOv5的密集行人检测算法[J].计算机工程, 2025, 51(3): 216-228.
HU Q, PI J Y, HU W C, et al. Research on dense pedestrian detection algorithm based on improved YOLOv5[J]. Computer Engineering, 2025, 51(3): 216-228.
[15] LI N F, BAI X L, SHEN X F, et al. Dense pedestrian detection based on GR-YOLO[J]. Sensors, 2024, 24(14): 4747.
[16] 徐振峰, 许云峰, 于子洲, 等. 针对密集行人检测任务中多尺度目标的检测算法[J/OL]. 计算机工程与应用, 2024: 1-15 (2024-09-12)[2024-09-20]. https://kns.cnki.net/kcms/detail/11.2127.TP.20240912.1036.014.html.
XU Z F, XU Y F, YU Z Z, et al. Multi-scale target detection algorithm for dense pedestrian detection task[J/OL]. Computer Engineering and Applications, 2024: 1-15 (2024-09-12) [2024-09-20]. https://kns.cnki.net/kcms/detail/11.2127.TP.20240912.1036.014.html.
[17] ZHONG J C, CHEN J Y, MIAN A. DualConv: dual convolutional kernels for lightweight deep neural networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2023, 34(11): 9528-9535.
[18] YANG L. ZHANG R Y LI L, et al. SimAM: a simple, parameter-free attention module for convolutional neural networks[C]//Proceedings of the International Conference on Machine Learning, 2021:11863-11874.
[19] YU Z P, HUANG H B, CHEN W J, et al. YOLO-FaceV2: a scale and occlusion aware face detector[J]. Pattern Recognition, 2024, 155: 110714.
[20] WANG A, CHEN H, LIN Z J, et al. Rep ViT: revisiting mobile CNN from ViT perspective[C]//Proceedings of the 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2024: 15909-15920.
[21] TAN M X, PANG R M, LE Q V. EfficientDet: scalable and efficient object detection[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 10778-10787.
[22] ZHANG X D, ZENG H, GUO S, et al. Efficient long-range attention network forImage super-resolution[C]//Proceedings of the European Conference on Computer Vision. Cham: Springer, 2022: 649-667.
[23] 肖振久, 孔祥旭, 宗佳旭, 等. 自适应聚焦损失的图像目标检测算法[J]. 计算机工程与应用, 2021, 57(23): 185-192.
XIAO Z J, KONG X X, ZONG J X, et al. Image object detection algorithm based on adaptive focal loss[J]. Computer Engineering and Applications, 2021, 57(23): 185-192.
[24] ZHANG H, ZHANG S. Shpae-IoU: more accurate metric considering bounding box shape and scale [J]. arXiv:2321.17663, 2017.
[25] ZHANG S F, XIE Y L, WAN J, et al. WiderPerson: a diverse dataset for dense pedestrian detection in the wild[J]. IEEE Transactions on Multimedia, 2020, 22(2): 380-393.
[26] ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: faster and better learning for bounding box regression[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 12993-13000.
[27] REZATOFIGHI H, TSOI N, GWAK J, et al. Generalized intersection over union: a metric and a loss for bounding box regression[C]//Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2019: 658-666.
[28] ZHANG Y F, REN W Q, ZHANG Z, et al. Focal and efficient IOU loss for accurate bounding box regression[J]. Neurocomputing, 2022, 506: 146-157.
[29] REDMON J, FARHADI A. YOLOv3: an incremental improvement[C]//Proceedings of the IEEE International Conference on Computer Vision and Pattem Recognition. Berlin, Heidelberg: Springer, 2018: 1-6.
[30] 喻伟,周劲,黄金良.基于改进YOLOv5稻米垩白检测的研究[J].中国粮油学报, 2025, 40(2): 41-48.
YU W, ZHOU J, HUANG J L. Research on rice chalkiness detection based on improved YOLOv5[J]. Journal of the Chinese Cereals and Oils, 2025, 40(2): 41-48.
[31] WANG C Y, 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 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2023: 7464-7475.
[32] WANG A, CHEN H, LIU L, et al YOLOv10: real-time end-to-end object detection[J]. arXiv:2405.14258, 2024.
[33] KHANAM R, HUSSAIN M. YOLOv11: an overview of the key architectural enhancements[J]. arXiv:2410.17725, 2024.
[34] 王泽宇, 徐慧英, 朱信忠, 等. 基于YOLOv8改进的密集行人检测算法: MER-YOLO[J]. 计算机工程与科学, 2024, 46(6): 1050-1062.
WANG Z Y, XU H Y, ZHU X Z, et al. An improved dense pedestrian detection algorithm based on YOLOv8: MER-YOLO[J]. Computer Engineering & Science, 2024, 46(6): 1050-1062.
[35] LIU Q L, YE H X, WANG S M, et al. YOLOv8-CB: dense pedestrian detection algorithm based on in-vehicle camera[J]. Electronics, 2024, 13(1): 236.
[36] XU Y X, WEN M, HE W, et al. An improved multi-scale and knowledge distillation method for efficient pedestrian detection in dense scenes[J]. Journal of Real-Time Image Processing, 2024, 21(4): 126.