基于二维和三维人体姿态估计的跌倒检测

李现国, 段显霖, 杨彦利, 蒋鹏, 张磊

天津工业大学学报 ›› 2026, Vol. 45 ›› Issue (4) : 80 -88.

PDF (2029KB)
天津工业大学学报 ›› 2026, Vol. 45 ›› Issue (4) : 80 -88. DOI: 10.3969/j.issn.1671-024x.2026.04.010
电子信息与自动化

基于二维和三维人体姿态估计的跌倒检测

    李现国1,2, 段显霖1, 杨彦利1,2, 蒋鹏3, 张磊3
作者信息 +

Fall detection based on 2D and 3D human pose estimation

    Li Xianguo1,2, Duan Xianlin1, Yang Yanli1,2, Jiang Peng3, Zhang Lei3
Author information +
文章历史 +
PDF (2077K)

摘要

针对现有跌倒检测技术普遍存在的误检漏检问题,提出联合二维和三维人体姿态估计的跌倒检测策略:基于Mamba模型设计Mamba-Attention注意力机制、改进C3_M模块,提出二维人体姿态估计算法MambaPose,提高二维人体关键点检测能力;融合Flash-Attention、DCT变换与低通滤波,提出三维人体姿态估计算法FlashPose,提高二维人体关键点映射到三维空间的准确性和实时性;将MambaPose、FlashPose和K近邻分类器(KNN)结合,构建出一种高准确率实时跌倒检测方法MFKF。公共数据集上的实验结果表明:在COCO数据集上,MambaPose算法与YOLOv7Pose相比,AP值提高了2.5个百分点,显著提升了二维人体关键点检测的准确性,而且小目标检测的置信度更高、处理低分辨率图像时鲁棒性更强;在Human3.6M数据集上,FlashPose算法的MPJPE和FPS值分别达到47.2 mm和314,优于PoseFormerV2,准确性更高、计算速度更快;在UR、Le2i和Multicam这3个跌倒检测数据集上,本文MFKF方法的F1、灵敏度、特异度、准确率均为最优,且FPS值均高于22.7,实现了跌倒的准确和实时检测。代码及演示视频链接:https://github.com/xgli411/MFKF.

Abstract

To address the problem of false positives and false negatives in the existing fall detection technologies, a fall detection strategy integrating two-dimensional and three-dimensional human pose estimation is proposed. Based on the Mamba model, the Mamba-Attention mechanism is designed, the C3_M module is enhanced, and the 2D human pose estimation algorithm MambaPose is introduced. By integrating Flash-Attention, DCT transform and low-pass filtering, the 3D human pose estimation algorithm FlashPose is proposed to improve the accuracy and real-time performance of mapping 2D human keypoints to 3D space. Finally, MFKF, a high accuracy real-time fall detection method, is constructed by combining MambaPose, FlashPose, and the KNN classifier. The experimental results on the public datasets show that: On the COCO dataset, the MambaPose algorithm improves the AP value by 2.5 percentage points compared with YOLOv7Pose, which significantly improves the accuracy of 2D human keypoint detection, and has higher confidence in small target detection and greater robustness when processing low-resolution images; on the Human3.6M dataset, the FlashPose algorithm′s MPJPE and FPS values reach 47.2 mm and 314, respectively, which are better than PoseFormerV2, with higher accuracy and faster computation speed; on the three fall detection datasets of UR, Le2i, and Multicam, the proposed MFKF method has the optimal F1, sensitivity, specificity, and accuracy indexes, and the FPS values are higher than 22.7, which realizes accurate and real-time detection of falls. Code and demonstration videos are available at https://github.com/xgli411/MFKF.

关键词

跌倒检测 / 二维人体姿态估计 / 三维人体姿态估计 / K近邻分类器(KNN) / Mamba模型

Key words

fall detection / 2D human pose estimation / 3D human pose estimation / KNN classifier / Mamba model

引用本文

引用格式 ▾
李现国, 段显霖, 杨彦利, 蒋鹏, 张磊. 基于二维和三维人体姿态估计的跌倒检测[J]. 天津工业大学学报, 2026, 45(4): 80-88 DOI:10.3969/j.issn.1671-024x.2026.04.010

登录浏览全文

4963

注册一个新账户 忘记密码

参考文献

[1] World Health Organization (WHO). Falls[EB/OL]. (2021-04-26)[2024-11-06]. https://www.who.int/news-room/fact-shee-ts/detail/falls.
[2] Burns E, Kakara R.Deaths from falls among persons aged ≥65 years: United States, 2007-2016[J]. MMWR Morbidity and Mortality Weekly Report, 2018, 67(18): 509-514.
[3] Kelsey J L, Procter-Gray E, Hannan M T, et al.Heterogeneity of falls among older adults: Implications for public health prevention[J]. American Journal of Public Health, 2012, 102(11): 2149-2156.
[4] Yu M, Rhuma A, Naqvi S M, et al.A posture recognition-based fall detection system for monitoring an elderly person in a smart home environment[J]. IEEE Transactions on Information Technology in Biomedicine, 2012, 16(6): 1274-1286.
[5] Lu N, Wu Y D, Feng L, et al.Deep learning for fall detection: Three-dimensional CNN combined with LSTM on video kinematic data[J]. IEEE Journal of Biomedical and Health Informatics, 2019, 23(1): 314-323.
[6] Chen W M, Jiang Z J, Guo H L, et al.Fall detection based on key points of human-skeleton using OpenPose[J]. Symmetry, 2020, 12(5): 744.
[7] Cao Z, Simon T, Wei S H, et al.Realtime multi-person 2D pose estimation using part affinity fields[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, HI, USA: IEEE, 2017: 1302-1310.
[8] Zhang Y, Gan J S, Zhao Z W, et al.A real-time fall detection model based on BlazePose and improved ST-GCN[J]. Journal of Real-Time Image Processing, 2023, 20(6): 121.
[9] Amsaprabhaa M, Nancy Jane Y, Khanna Nehemiah H.Multimodal spatiotemporal skeletal kinematic gait feature fusion for vision-based fall detection[J]. Expert Systems with Applications, 2023, 212: 118681.
[10] Beddiar D R, Oussalah M, Nini B.Fall detection using body geometry and human pose estimation in video sequences[J]. Journal of Visual Communication and Image Representation, 2022, 82: 103407.
[11] Kan X, Zhu S H, Zhang Y H, et al.A lightweight human fall detection network[J]. Sensors, 2023, 23(22): 9069.
[12] Núñez-Marcos A, Arganda-Carreras I.Transformer-based fall detection in videos[J]. Engineering Applications of Artificial Intelligence, 2024, 132: 107937.
[13] 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]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver, BC, Canada:IEEE, 2023: 7464-7475.
[14] Gu A, Dao T. Mamba: Linear-time sequence modeling with selective state spaces[PP/OL]. V2. arXiv (2024-05-31)[2024-11-06]. https://doi.org/10.48550/arXiv.2312.00752.
[15] Sunkara R, Luo T.No more strided convolutions or pooling: A new CNN building block for low-resolution images and small objects[C]//Machine Learning and Knowledge Discovery in Databases. Cham: Springer Nature Switzerland, 2023: 443-459.
[16] Zhao Q T, Zheng C, Liu M Y, et al.PoseFormerV2: Exploring frequency domain for efficient and robust 3D human pose estimation[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver, BC, Canada:IEEE, 2023: 8877-8886.
[17] Dao T, Fu D, Ermon S, et al.FlashAttention: Fast and memory-efficient exact attention with IO-awareness[C]//Advances in Neural Information Processing Systems 35. New Orleans, Louisiana, USA: Neural Information Processing Systems Foundation, Inc.(NeurIPS), 2022: 16344-16359.
[18] Lin T Y, Maire M, Belongie S, et al.Microsoft COCO: Common objects in context[C]//Computer Vision-ECCV 2014. Cham: Springer, 2014: 740-755.
[19] Maji D, Nagori S, Mathew M, et al.YOLO-pose: Enhancing YOLO for multi person pose estimation using object keypoint similarity loss[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). New Orleans, LA, USA: IEEE, 2022: 2636-2645.
[20] Ionescu C, Papava D, Olaru V, et al.Human3.6M: Large scale datasets and predictive methods for 3D human sensing in natural environments[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 36(7): 1325-1339.
[21] Kwolek B, Kepski M.Human fall detection on embedded platform using depth maps and wireless accelerometer[J]. Computer Methods and Programs in Biomedicine, 2014, 117(3): 489-501.
[22] Charfi I, Miteran J, Dubois J, et al.Optimized spatio-temporal descriptors for real-time fall detection: Comparison of support vector machine and AdaBoost-based classification[J]. Journal of Electronic Imaging, 2013, 22(4): 041106.
[23] Tian Z, Chen H, Shen C H.DirectPose: Direct end-to-end multi-person pose estimation[PP/OL]. V2. arXiv (2019-11-24)[2024-11-06]. https://doi.org/10.48550/arXiv.1911.07451.
[24] Mao W A, Tian Z, Wang X L, et al.FCPose: Fully convolutional multi-person pose estimation with dynamic instance-aware convolutions[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, TN, USA: IEEE, 2021: 9030-9039.
[25] Shi D H, Wei X, Yu X D, et al.InsPose: Instance-aware networks for single-stage multi-person pose estimation[C]//Proceedings of the 29th ACM International Conference on Multimedia. New York: ACM, 2021: 3079-3087.
[26] Shi D H, Wei X, Li L Q, et al.End-to-end multi-person pose estimation with transformers[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New Orleans, LA, USA: IEEE, 2022: 11059-11068.
[27] Wang D K, Zhang S L.Contextual instance decoupling for robust multi-person pose estimation[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New Orleans, LA, USA: IEEE, 2022: 11050-11058.
[28] Yang J, Zeng A, Liu S, et al.Explicit box detection unifies end-to-end multi-person pose estimation[C]//2023 Interna-tional Conference on Learning Representations (ICLR). Kigali, Rwanda: IEEE, 2023: 598-607.
[29] McNally W, Vats K, Wong A, et al. Rethinking keypoint representations: Modeling keypoints and poses as objects for multi-person human pose estimation[C]//Computer Vision-ECCV 2022. Cham: Springer Nature Switzerland, 2022: 37-54.
[30] Li W H, Liu H, Tang H, et al.MHFormer: Multi-hypothesis transformer for 3D human pose estimation[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New Orleans, LA, USA: IEEE, 2022: 13137-13146.
[31] Zheng C, Zhu S J, Mendieta M, et al.3D human pose estimation with spatial and temporal transformers[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV). Montreal, QC, Canada: IEEE, 2021: 11636-11645.
[32] Egawa R, Miah A S M, Hirooka K, et al. Dynamic fall detection using graph-based spatial temporal convolution and attention network[J]. Electronics, 2023, 12(15): 3234.
[33] Fei K X, Wang C, Zhang J X, et al.Flow-pose Net: An effective two-stream network for fall detection[J]. The Visual Computer, 2023, 39(6): 2305-2320.
[34] Li S Y, Song X, Xu S Y, et al.Dilated spatial-temporal convolutional auto-encoders for human fall detection in surveillance videos[J]. ICT Express, 2023, 9(4): 734-740.

基金资助

天津市技术创新引导专项(23YDTPJC00290)

AI Summary AI Mindmap
PDF (2029KB)

0

访问

0

被引

详细

导航
相关文章

AI思维导图

/

〈 〉