基于检测增强型YOLOv3-tiny的道路场景行人检测

doi:10.3969/j.issn.1671-7775.2024.04.010

摘要
图/表
参考文献(0)
相关文章 (15)

全文: PDF (5936 KB) HTML (1 KB)
输出: BibTeX | EndNote (RIS)

摘要为了给驾驶员提供实时准确的行人信息、减少交通事故的发生，提出一种检测增强型YOLOv3-tiny（detection of enhanced YOLOv3-tiny，DOEYT）行人检测算法.创建鲁棒的特征提取网络，首先使用非对称最大池化进行下采样，防止随着感受野增大行人横向特征的丢失；其次使用Hardswish作为卷积层的激活函数优化网络性能；最后使用GC（globe context）自注意力机制获得全文特征信息.在分类回归网络部分，采用三尺度检测策略，提升小尺度行人目标的检测精度；使用k-means++算法重新生成数据集锚框，提高网络收敛速度.构建行人检测数据集并分为训练集和测试集，对DOEYT算法的性能进行试验验证.结果表明，非对称最大池化、Hardswish函数、GC自注意力机制分别使平均准确率AP提高14.4%、7.9%、10.8%；DOEYT算法在测试集上检测的平均准确率高达91.2%，检测速度为103帧/s，可见该算法可快速准确地检测行人，降低交通事故发生的风险.

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	田亮1
	2
	金积德1
	2
	郑庆祥1
	2

关键词 ：行人检测, 深度学习, 卷积神经网络, 非对称最大池化, 激活函数, 自注意力机制, 多尺度检测, YOLOv3-tiny

Abstract：To provide drivers with real-time and accurate pedestrian information and reduce traffic accidents, the detection of enhanced YOLOv3-tiny (DOEYT) pedestrian detection algorithm was proposed. The robust feature extraction network was established, and the asymmetric max-pooling was used for down sampling to prevent the loss of lateral pedestrian features due to the increased receptive field. Hardswish was employed as activation function for the convolutional layers to optimize network performance, and the global context (GC) self-attention mechanism was used to capture holistic feature information. In the classification and regression network, the three-scale detection strategy was adopted to improve the accuracy of small-scale pedestrian target detection. The k-means++ algorithm was used to regenerate dataset anchor boxes for enhancing network convergence speed. The pedestrian detection dataset was constructed and divided into training and testing sets to evaluate DOEYT performance. The results show that by the asymmetric max-pooling, Hardswish function and GC self-attention mechanism, AP values are increased by 14.4%, 7.9% and 10.8%, respectively. On the testing set, DOEYT achieves average precision of 91.2% and detection speed of 103 frames per second, which demonstrates that the proposed algorithm can quickly and accurately detect pedestrians for reducing the risk of traffic accidents.

Key words： pedestrian detection')" href="#">

pedestrian detection deep learning convolutional neural networks asymmetric max-pooling activation function self-attention mechanism multi-scale detection

')" href="#">YOLOv3-tiny

收稿日期: 2022-05-12

基金资助:“211工程”专项资金资助项目(40120411); 高等学校学科创新引智计划项目（B17034）; 教育部创新研究团队发展计划项目(IRT_17R83)

作者简介: 田亮(1998—),男,河北唐山人,硕士研究生(aries_tian98@163.com),主要从事人工智能计算机视觉研究. 郑庆祥(1970—),男,中国台湾人,副教授(通信作者,chcheng168@whut.edu.cn),主要从事芯片微流散热与纳米层级动力电池等研究.

引用本文:

田亮1,2，金积德1,2，郑庆祥1,2. 基于检测增强型YOLOv3-tiny的道路场景行人检测[J]. 江苏大学学报（自然科学版）, 2024, 45(4): 441-448. TIAN Liang1, 2， JIN Jide1,2， ZHENG Qingxiang1,2. Road scene pedestrian detection based on detection-enhanced YOLOv3-tiny[J]. Journal of Jiangsu University(Natural Science Eidtion) , 2024, 45(4): 441-448.

链接本文:

http://zzs.ujs.edu.cn/xbzkb/CN/ 10.3969/j.issn.1671-7775.2024.04.010 或 http://zzs.ujs.edu.cn/xbzkb/CN/Y2024/V45/I4/441

［1］	KUMAR K, MISHRA R K. A heuristic SVM based pedestrian detection approach employing shape and texture descriptors[J]. Multimedia Tools and Applications, 2020,79(29/30):21389-21408.
［2］	WEI X, LU W, BAO P, et al. MGA for feature weight learning in SVM: a novel optimization method in pedestrian detection[J]. Multimedia Tools and Applications, 2018,77(7):9021-9037.
［3］	尧佼,于凤芹. 基于候选区域定位与HOG-CLBP特征组合的行人检测[J]. 激光与光电子学进展, DOI:10.3788/LOP202158.0210015.
	YAO J, YU F Q. Pedestrian detection based on combination of candidate region location and HOG-CLBP features[J]. Advances in Laser and Optoelectronics Progress, DOI:10.3788/LOP202158.0210015.(in Chinese)
［4］	王彬,李小曼,赵作鹏. 基于改进Faster RCNN的驾驶员手持通话检测[J]. 江苏大学学报(自然科学版), 2023,44(3):318-323.
	WANG B, LI X M, ZHAO Z P.Hand-held call detection of driver based on improved Faster RCNN[J]. Journal of Jiangsu University (Natural Science Edition), 2023,44(3):318-323.(in Chinese)
［5］	欧阳继红,王梓明,刘思光. 改进多尺度特征的YOLO_v4目标检测方法[J]. 吉林大学学报(理学版), 2022,60(6):1349-1355.
	OUYANG J H, WANG Z M, LIU S G. YOLO_v4 object detection method with improved multi-scale features[J]. Journal of Jilin University (Science Edition), 2022,60(6):1349-1355.(in Chinese)
［6］	王卜,何扬. 基于改进YOLOv3的交通标志检测[J]. 四川大学学报(自然科学版), DOI: 10.19907/j.0490-6756.2022.012004.
	WANG B, HE Y. Traffic sign detection based on improved YOLOv3[J]. Journal of Sichuan University (Natural Science Edition), DOI:10.19907/j.0490-6756.2022.012004.(in Chinese)
［7］	夏英,黄秉坤. 采用改进YOLOv3的高分辨率遥感图像目标检测[J]. 重庆邮电大学学报(自然科学版), 2022,34(3):383-392.
	XIA Y, HUANG B K. Object detection of high resolution remote sensing images based on improved YOLOv3[J]. Journal of Chongqing University of Posts and Telecommunications (Natural Science Edition), 2022,34(3):383-392.(in Chinese)
［8］	方阳,刘英杰,孙立博,等. 基于SSD模型的人脸检测与头部姿态估计融合算法[J]. 江苏大学学报(自然科学版), 2019,40(4):451-457.
	FANG Y, LIU Y J, SUN L B, et al. Fusion algorithm of face detection and head pose estimation based on SSD model[J]. Journal of Jiangsu University (Natural Science Edition), 2019,40(4):451-457.(in Chinese)
［9］	HSU W Y, LIN W Y. Adaptive fusion of multi-scale YOLO for pedestrian detection[J]. IEEE Access, 2021(9):110063-110073.
［10］	ZHENG Y, IZZAT I H, ZIAEE S. GFD-SSD: gated fusion double SSD for multispectral pedestrian detection[J]. arXiv preprint arXiv:1903.06999v2.
［11］	KIM J H, BATCHULUUN G, PARK K R. Pedestrian detection based on faster R-CNN in nighttime by fusing deep convolutional features of successive images[J]. Expert Systems with Applications, 2018,114:15-33.
［12］	CAO Y, XU J R, LIN S, et al. GCNet: non-local networks meet squeeze-excitation networks and beyond[C]∥Proceedings of the 17th IEEE/CVF International Conference on Computer Vision Workshop.Piscataway：IEEE, DOI:10.1109/ICCVW.2019.00246.
［13］	HOWARD A, SANDLER M, CHU G, et al. Searching for mobilenetV3[C]∥Proceedings of the 17th IEEE/CVF International Conference on Computer Vision Workshop.Piscataway：IEEE, DOI:10.1109/ICCV.2019.00140.