Lightweight Non-Motor Vehicle Target Detection Based on Re-Parameterization

doi:10.3778/j.issn.1002-8331.2403-0292

Abstract

Abstract: Aiming at the problems that the number of targets in non-motor vehicle detection is easy to be occluded, the number of detection model parameters is large and the detection speed is slow, combining YOLOv8 algorithm and HGNetV2 network, a lightweight non-motor vehicle target detection algorithm based on re-parameterization is proposed, called RCH-YOLO. Firstly, the improved re-parameterized backbone network based on HGNetV2 is used to reduce the number of parameters and calculations and improve the detection speed. Sceondly, Slim-neck is used to improve the neck network, and TGConv convolution is used in the detector to reduce the complexity of the network while maintaining accuracy. Finally, the Wise-IoU loss function is used to calculate the positioning loss, accelerate the convergence speed of the model, and improve the recognition accuracy of overlapping targets. Experimental results show that compared with the baseline model, the average detection accuracy of the proposed algorithm is improved by 0.3 percentage points, and the number of parameters, model weight and GFLOPs are reduced by 47.9%, 44.2% and 54.9% respectively, which verifies the effectiveness of the proposed algorithm.

Key words: lightweight, non-motor vehicle, YOLOv8, re-parameterization, HGNetV2

摘要： 针对非机动车检测中目标数量多易被遮挡、检测模型参数量较大且检测速度慢的问题，结合YOLOv8算法和HGNetV2网络,提出一种基于重参数化的轻量化非机动车目标检测算法RCH-YOLO。使用基于HGNetV2改进的重参数化骨干网络，减少参数量和计算量，提高检测速度;在此基础上，使用Slim-neck改进颈部网络，并在检测器中使用TGConv卷积，降低网络复杂度的同时保持准确性；采用Wise-IoU损失函数计算定位损失，加速模型收敛速度，提高对重叠目标的识别准确率。实验结果显示，相较于基准模型，RCH-YOLO算法平均检测精度提高了0.3个百分点，参数量、模型大小和GFLOPs分别降低了47.9%、44.2%和54.9%，验证了其有效性。

关键词: 轻量化, 非机动车, YOLOv8, 重参数化, HGNetV2

MA Chaofan, LI Xiang, WANG Xiaoxia, CHEN Xiao. Lightweight Non-Motor Vehicle Target Detection Based on Re-Parameterization[J]. Computer Engineering and Applications, 2024, 60(19): 190-198.

马超凡, 李翔, 王晓霞, 陈晓. 基于重参数化的轻量化非机动车目标检测[J]. 计算机工程与应用, 2024, 60(19): 190-198.

References

[1] QIAN Y N, SUN Q N, FEI G Q. Riding behavior and electric bike traffic crashes: a Chinese case-control study[J]. Traffic Injury Prevention, 2020, 21(2): 24-28.
[2] REN S, HE K, 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.
[3] HE K, GKIOXARI G, DOLLAR P, et al. Mask R-CNN[C]//Proceedings of the 2017 IEEE International Conference on Computer Vision (ICCV), 2017: 2980-2988.
[4] 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 (CVPR), 2016: 779-788.
[5] LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]//Proceedings of the European Conference on Computer Vision(ECCV), 2016: 21-37.
[6] LIN TY, MAIRE M, BELONGIE S, et al. Microsoft COCO: common objects in context[C]//Proceedings of the European Conference on Computer Vision, 2014: 740-755.
[7] EVERINGHAM M, ESLAMI S, VAN G, et al. The PASCAL visual object classes challenge: a retrospective[J]. International Journal of Computer Vision, 2015, 111(1): 98-136.
[8] WANG X, XIAO T, JIANG Y, et al. Repulsion loss: detecting pedestrians in a crowd[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 7774-7783.
[9] ZHANG S F, WEN L Y, BIAN X, et al. Occlusion-aware R-CNN: detecting pedestrians in a crowd[C]//Proceedings of the European Conference on Computer Vision (ECCV), 2018: 657-674.
[10] 叶佳林, 苏子毅, 马浩炎, 等. 改进YOLOv3的非机动车检测与识别方法[J].计算机工程与应用, 2021, 57(1):194-199.
YE J L, SU Z Y, MA H Y, et al. Improved YOLOv3 non-motor vehicles detection and recognition method[J]. Computer Engineering and Applications, 2021, 57(1): 194-199.
[11] 邵奇可, 李路, 周宇, 等. 一种基于滑动窗口优化算法的行人检测算法[J]. 浙江工业大学学报, 2015, 43(2): 212-216.
SHAO Q K, LI L, ZHOU Y, et al. Pedestrian detection in videos based on optimization algorithm using sliding window[J]. Journal of Zhejiang University of Technology, 2015, 43(2): 212-216.
[12] HAMID R, NATHAN T,GWAK J Y, 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 (CVPR), 2019: 658-666.
[13] 周辉, 王维莉. 基于无人机视频的非机动车道交通冲突分析[J]. 计算机工程与应用, 2023, 59(16): 324-329.
ZHOU H, WANG W L. Traffic conflict analysis of non-motor vehicle lanes based on UAV video[J]. Computer Engineering and Applications, 2023, 59(16): 324-329.
[14] HOWARD A , ZHU M, CHEN B, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[J]. arXiv:1704.04861, 2017.
[15] ZHANG X, ZHOU X, LIN M, et al. ShuffleNet: an extremely efficient convolutional neural network for mobile devices[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 6848-6856.
[16] HAN K, WANG Y, TIAN Q, et al. GhostNet: more features from cheap operations[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020: 1577-1586.
[17] REIS D, KUPEC J, HONG J, et al.Real-time flying object detection with YOLOv8[J]. arXiv:2305.09972, 2023.
[18] LI H, LI J, WEI H, et al. Slim-neck by GSConv: a better design paradigm of detector architectures for autonomous vehicles[J]. arXiv:2206.02424, 2022.
[19] YANI I, DUNCAN R, ROBERTO C, et al. Deep roots: improving CNN efficiency with hierarchical filter groups[J]. arXiv:1605.06489v3, 2016.
[20] TONG Z, CHEN Y, XU Z, et al. Wise-IoU: bounding box regression loss with dynamic focusing mechanism[J]. arXiv:2301.10051, 2023.
[21] DING X, ZHANG X, MA N, et al. RepVGG: making VGG-style ConvNets great again[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021: 13728-13737.
[22] ZHENG Z, WANG P, LIU W, et al. Distance-IoU loss: faster and better learning for bounding box regression[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2020: 12993-13000.