|Table of Contents|

Improved YOLOv11-Pose-based recognition method for pedestrian improper crossing posture(PDF)

长安大学学报（自然科学版）[ISSN:1006-6977/CN:61-1281/TN]

Issue:
2026年2期
Page:
117-127
Research Field:
交通工程
Publishing date:

Info

Title:
Improved YOLOv11-Pose-based recognition method for pedestrian improper crossing posture
Author(s):
PEI Yu-long;  CONG Wei
(School of Civil Engineering and Transportation, Northeast Forestry University, Harbin 150040, China)
Keywords:
traffic engineering;  posture recognition;  YOLOV11-Pose;  improper crossing;  WIoU;  self-calibration module
PACS:
U491.1
DOI:
10.19721/j.cnki.1671-8879.2026.02.009
Abstract:
Aiming at the problem that pedestrian improper crossing postures are difficult to accurately recognize in complex traffic scenes, a pedestrian improper crossing posture recognition algorithm based on the improved YOLOv11-Pose model was proposed to achieve accurate detection of dangerous crossing postures. Systematic optimization was conducted on the model architecture and loss function. First, the WIoU loss function was introduced to replace the original bounding box regression loss, and a dynamic weight adjustment mechanism was adopted to enhance the learning capability for hard samples, thereby improving localization accuracy and accelerating model convergence. Second, the SBAM attention mechanism was embedded into the backbone network, constructing a spatial-channel collaborative enhancement structure to improve the feature representation capability for long-distance and small-scale pedestrians. Furthermore, the self-calibration module was integrated with the C3k2 module, strengthening feature re-calibration and boundary-region detail perception, which improved the stability of keypoint detection and the accuracy of improper crossing posture discrimination under complex backgrounds. Based on human keypoint information, the model constructs posture structural feature mappings to achieve fine-grained recognition of improper crossing postures. Experimental results on a self-built pedestrian improper crossing posture dataset show that the improved model increases mAP@0.5 by 4.2%, precision by 3.4%, and recall by 6.7% compared with the original model. In the small-target and complex-scene subsets, the miss detection rate decreases significantly, the number of convergence epochs is reduced, and the training process becomes more stable. The proposed method demonstrates significant improvements in detection accuracy, robustness, and generalization capability. It can serve as a key perception module in intelligent driving assistance and public safety monitoring systems, providing support for real-time recognition and active warning of dangerous road-crossing behaviors. It has practical value for reducing traffic accident rates and improving road safety and traffic operation efficiency.5 tabs, 6 figs, 33 refs.

References:

[1] 尹 岩,马社强.过街行人的交通特性及对策研究[J].交通科技与经济,2013,15(6):18-22,26.
YIN Yan, MA She-qiang. Research on traffic characteristics and countermeasures of cross-street pedestrians[J]. Transportation Science and Economy, 2013, 15(6): 18-22,26.
[2]何永明,尚庆鹏,王 爽.城市行人过街交通安全警示系统研究[J].交通科技与经济,2018,20(4):9-12.
HE Yong-ming, SHANG Qing-peng, WANG Shuang. Research on traffic safety warning system for urban pedestrian crossing[J]. Transportation Science and Economy, 2018, 20(4): 9-12.
[3]姜 辉,李 珊,黄扬翔,等.基于图像处理的行人过街安全警示系统设计与研究[J].现代信息科技,2024,8(12):96-99,104.
JIANG Hui, LI Shan, HUANG Yang-xiang, et al. Design and research of pedestrian crossing safety warning system based on image processing[J]. Modern Information Technology, 2024, 8(12): 96-99, 104.
[4]王诗瑶.基于车载摄像头的行人检测与预警方法研究[D].大连:大连海事大学,2023.
WANG Shi-yao. Research on pedestrian detection and early warning method based on vehicle camera[D]. Dalian: Dalian Maritime University, 2023.
[5]娄 文,郭杜杜,张 杰,等.基于YOLOv7的驾驶人使用手机与抽烟行为识别方法[J].电子测量技术,2023,46(21):123-131.
LOU Wen, GUO Du-du, ZHANG Jie, et al. YOLOv7-based recognition method for drivers' mobile phone use and smoking behaviour[J]. Electronic Measurement Technology, 2023, 46(21): 123-131.
[6]李千帆,张彦会,谢鹏鹏,等.基于改进YOLOv8的雨天行人检测方法[J/OL].广西科技大学学报,2026.https://link.cnki.net/urlid/45.1395.t.20241128.20
39.004.
LI Qian-fan, ZHANG Yan-hui, XIE Peng-peng, et al. Pedestrian detection method in rainy days based on improved YOLOv8[J/OL]. Journal of Guangxi University of Science and Technology, 2026.https://link.cnki.net/urlid/45.1395.t.20241128.2039.004.
[7]魏 志,刘 罡,张 旭.基于MobileNet的轻量化密集行人检测算法[J].软件工程,2024,27(6):6-9.
WEI Zhi, LIU Gang, ZHANG Xu. Lightweight dense pedestrian detection algorithm based on MobileNet[J]. Software Engineering, 2024, 27(6): 6-9.
[8]王世芳,徐 琨,陈明瑶.一种基于特征金字塔的快速行人检测方法[J].长安大学学报(自然科学版),2018,38(5):231-237.
WANG Shi-fang, XU Kun, CHEN Ming-yao. A fast pedestrian detection method based on feature pyramid[J]. Journal of Chang'an University(Natural Science Edition), 2018, 38(5): 231-237.
[9]刘 赏,代 娆,周煜炜,等.增强人体关键点特征的姿态估计算法[J/OL].计算机辅助设计与图形学学报,2024,DOI:10.3724/SP.J.1089.2023-00706.
LIU Shang, DAI Rao, ZHOU Yu-wei, et al. Enhanced pose estimation algorithm for human keypoint features[J/OL]. Journal of Computer-Aided Design and Graphics, 2024,DOI: 10.3724/SP.J.1089.2023-00706.
[10]柏 强,邵宇麒,蒙思源,等. 基于视频的机场出发层违规接客车辆识别方法[J]. 长安大学学报(自然科学版), 2022,42(4):73-86.
BAI Qiang, SHAO Yu-ki, MONG Si-yuan, et al. A video-based method for identifying non-compliant pick-up vehicles on the departure level of airports[J]. Journal of Chang'an University(Natural Science Edition), 2022, 42(4): 73-86.
[11]马明旭,马 宏,宋华伟.基于YOLO-Pose的城市街景小目标行人姿态估计算法[J].计算机工程,2024,50(4):177-186.
MA Ming-xu, MA Hong, SONG Hua-wei. YOLO-Pose-based pedestrian pose estimation algorithm for small targets in urban streetscape[J]. Computer Engineering, 2024, 50(4): 177-186.
[12]MAJI D, NAGORI S, MATHEWM, et al. YOLO-Pose: Enhancing YOLO for multi person pose estimation using object keypoint similarity loss[C]//IEEE. Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2022: 2637-2646.
[13]TAN M X, PANG R M, LEQ V. EfficientDet: scalable and efficient object detection[C]//IEEE. Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2020: 10781-10790.
[14]周薇娜,丁豪文,周 颖.一种海上弱小运动船舶实时检测方法[J].合肥工业大学学报(自然科学版),2021,44(9):1187-1192.
ZHOU Wei-na, DING Hao-wen, ZHOU Ying. A real-time detection method for weakly moving ships at sea[J]. Journal of Hefei University of Technology(Natural Science Edition), 2021, 44(9): 1187-1192.
[15]袁 莹. 基于深度图像的独居老人姿态动作识别研究[D]. 南宁:广西民族大学, 2023.
YUAN Ying. Research on gesture-action recognition of elderly living alone based on deep image[D]. Nanning: Guangxi University for Nationalities, 2023.
[16]TONG Z, CHEN Y, XU Z, et al. Wise-IoU: Bounding box regression loss with dynamic focusing mechanism[EB/OL]. [2025-02-01].https://arxiv.org/abs/2301.10051.
[17]NI Z, CHEN X, ZHAI Y, et al. Context-guided spatial feature reconstruction for efficient semantic segmentation[EB/OL]. [2024-05-10].https://doi.org/10.48550/arXiv.2405.06228.
[18]于昆平.基于计算机视觉的人体坐姿识别方法研究[D].吉林:吉林大学,2024.
YU Kun-ping. Research on human sitting posture recognition method based on computer vision[D]. Jilin: Jilin University, 2024.
[19]郭泰龙.航站楼旅客异常行为识别算法研究[D].太原:中北大学,2024.
GUO Tai-long. Research on abnormal behavior recognition algorithm for passengers in terminal building[D]. Taiyuan: North Central University, 2024.
[20]周 亮.基于监控视频的对象行为识别与交通流量预测[D].北京:北京邮电大学,2020.
ZHOU Liang. Object behavior recognition and traffic flow prediction based on surveillance video[D]. Beijing: Beijing University of Posts and Telecommunications, 2020.
[21]TANG H Y, TAN S H, SU T Y, et al. Upper body posture recognition using inertial sensors and recurrent neural networks[J]. Applied Sciences, 2021, 11(24): 12101.
[22]FENG L, LI Z, LIU C, et al. SitR: Sitting posture recognition using RF signals[J]. IEEE Internet of Things Journal, 2020, 7(12): 11492-11504.
[23]王 丽.基于深度学习的行人过街意图识别技术研究[D].重庆:重庆交通大学,2023.
WANG Li. Research on pedestrian crossing intention recognition technology based on deep learning[D]. Chongqing: Chongqing Jiaotong University, 2023.
[24]孙剑明,韩生权,沈子成,等.基于双卷积链的双目人体姿态距离定位识别[J].兵工学报,2022,43(11):2846-2854.
SUN Jian-ming, HAN Sheng-quan, SHEN Zi-cheng, et al. Binocular human posture distance localization recognition based on double convolutional chain[J]. Journal of Military Engineering, 2022, 43(11): 2846-2854.
[25]CHENG B, XIAO B, WANG J, et al. Higherhrnet: Scale-aware rep resentation learning for bottom-up human pose estimation[C]//IEEE. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. New York: IEEE, 2020: 5386-5395.
[26]崔家礼,刘永基,李子贺,等.轻量且高精度的姿态检测改进网络HG-YOLO[J/OL].计算机应用,2026.https://link.cnki.net/urlid/51.1307.TP.20250427.1449.006.
CUI Jia-li, LIU Yong-ji, LI Zi-he, et al. Lightweight and high-precision attitude detection improved network HG-YOLO[J/OL].Computer Applications, 2026.https://link.cnki.net/urlid/51.1307.TP.2025
0427.1449.006.
[27]SHAH M S.基于改进的YOLO架构、Inception-Resnet-V2和XGazeNet的司机(面部、头部姿态和眼球注视)监控系统[D].西安:长安大学,2024.
SHAH M S. Driver(face, head posture and eye gaze)monitoring system based on improved YOLO architecture, Inception-Resnet-V2 and XGazeNet[D]. Xi'an: Chang'an University, 2024.
[28]邵煜潇,鲁 涛,王震宇,等.结合多尺度大核卷积的红外图像人体检测算法[J/OL].智能系统学报,2025,DOI:10.11992/tis.202404027.
SHAO Yu-xiao, LU Tao, WANG Zhen-yu, et al. Human detection algorithm for infrared image combined with multi-scale large kernel convolution[J/OL]. Chinese Journal of Intelligent Systems, 2025,DOI: 10.11992/tis.202404027.
[29]WU J, CHEN M. Pedestrian pose estimation based on YOLO-swintransformer hybrid model[J]. World Electric Vehicle Journal, 2025, 16(12): 658.
[30]伍智华,程江华,刘 通,等.激光透窗低质量图像人体姿态识别技术研究[J].中国激光,2025,52(6):268-279.
WU Zhi-hua, CHENG Jiang-hua, LIU Tong, et al. Research on human pose recognition technology for low-quality laser window-penetrating images[J]. Chinese Journal of Lasers, 2025, 52(6): 268-279.
[31]尹秋燕,丁 婧,聂志刚.YOLO-AirPose:无人机航拍视角下的人体姿态估计算法[J/OL].计算机应用,2026.https://link.cnki.net/urlid/51.1307.tp.20250
929.1700.014.
YIN Qiu-yan, DING Jing, NIE Zhi-gang. YOLO-AirPose: A human pose estimation algorithm from UAV aerial perspective[J/OL]. Computer Applications, 2026,https://link.cnki.net/urlid/51.1307.tp.20250929.
1700.014.
[32]孟超月,房志斌,李聪聪,等.ELDMPose-YOLO:健身运动场景下姿态识别方法[J/OL].计算机工程与应用,2026,https://link.cnki.net/urlid/11.2127.TP.
20250928.1057.010.
MENG Chao-yue, FANG Zhi-bin, LI Cong-cong, et al. ELDMPose-YOLO: A pose recognition method for fitness exercise scenarios[J/OL]. Computer Engineering and Applications, 2026,https://link.cnki.net/urlid/11.2127.TP.20250928.1057.010.
[33]赵婉月.基于YOLOv5的目标检测算法研究[D].西安:西安电子科技大学,2021.
ZHAO Wan-yue. Research on target detection algorithm based on YOLOv5[D]. Xi'an: Xi'an Electronic Science and Technology University, 2021.

Memo

Memo:
-
Last Update: 2026-04-20