24, May 2026

Real-Time Human Posture and Action Recognition on Edge Devices Using Lightweight Pose Estimation

VOLUME-12; ISSUE-5; MAY-2026

Author(s): 1. Nityam Upadhyay, 2. Ms. Hetal Chokshi

Authors Affiliations:

  1. Nityam Upadhyay, Dept. of Computer Science & Engineering, Drs. Kiran & Pallavi Patel Global University (KPGU), Vadodara, India
  2. Ms. Hetal Choksi, Dept. of Computer Science & Engineering, Drs. Kiran & Pallavi Patel Global University (KPGU), Vadodara, India

DOIs:10.2015/IJIRMF/202605009     |     Paper ID: IJIRMF202605009


Abstract
Keywords
Cite this Article/Paper as
References

This paper presents a highly efficient, real-time Human Action Recognition system optimized for edge computing environments. Traditional state-of-the-art models for action recognition rely on computationally expensive pixel-dense video analysis, which necessitates high-end graphical processing units. This limitation prevents widespread deployment in consumer hardware and mobile surveillance systems. To overcome this computational bottleneck, we propose a decoupled, two-stage hybrid pipeline. In the first stage, spatial features are extracted using a MobileNetV2-backed pose estimator, which rapidly identifies key anatomical joints and converts dense video frames into sparse coordinate arrays. In the second stage, these normalized skeletal arrays are passed to a Deep Neural Network optimized for sequential posture classification. The proposed model is evaluated on a custom dataset consisting of four primary activities: sitting, standing, walking, and sleeping. Experimental results demonstrate that our lightweight architecture achieves a baseline test accuracy of eighty-two percent while maintaining an inference speed exceeding twenty-five frames per second on standard central processing unit hardware. These findings validate the efficacy of sparse spatial modeling for real-time applications in resource-constrained environments.

human action recognition, pose estimation, deep learning, edge computing, computer vision, spatial modeling

1. Nityam Upadhyay, 2. Ms. Hetal Chokshi (2026); Real-Time Human Posture and Action Recognition on Edge Devices Using Lightweight Pose Estimation, International Journal for Innovative Research in Multidisciplinary Field, ISSN(O): 2455-0620, Vol-12, Issue-5, Available on –   https://www.ijirmf.com/

  • Carreira and A. Zisserman, “Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset,” 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 2017, pp. 4724-4733.
  • Feichtenhofer, H. Fan, J. Malik, and K. He, “SlowFast Networks for Video Recognition,” 2019 IEEE/CVF International Conference on Computer Vision (ICCV), Seoul, Korea (South), 2019, pp. 6201-6210.
  • Arnab, M. Dehghani, G. Borgefors, C. Schmid, and T. Brox, “ViViT: A Video Vision Transformer,” 2021 IEEE/CVF International Conference on Computer Vision (ICCV), Montreal, QC, Canada, 2021, pp. 6816-6826.
  • Yan, Y. Xiong, and D. Lin, “Spatial Temporal Graph Convolutional Networks for Skeleton-Based Action Recognition,” Thirty-Second AAAI Conference on Artificial Intelligence, New Orleans, LA, USA, 2018.
  • M. Noori, et al., “A Robust Human Activity Recognition Approach Using OpenPose, Motion Heuristics, and Deep Learning,” IEEE Access, vol. 7, pp. 8894452-8894467, 2019.
  • Zheng, W. Wu, T. Yang, S. Zhu, C. Chen, R. Liu, J. Shen, N. Kehtarnavaz, and M. Shah, “Deep Learning-Based Human Pose Estimation: A Survey,” ACM Computing Surveys, vol. 56, no. 1, pp. 1-35, 2023.
  • Eason, B. Noble, and I. N. Sneddon, “On certain integrals of Lipschitz-Hankel type involving products of Bessel functions,” Phil. Trans. Roy. Soc. London, vol. A247, pp. 529–551, April 1955.

 

Post Views: 84

Download Full Paper

Download PDF No. of Downloads:8 | No. of Views: 66