Human Action Recognition

Model-architecture codesign for skeleton-based human action recognition

This project focuses on designing a real-time and hardware-efficient processor for skeleton-based action recognition using lightweight convolutional neural networks. The system targets embedded and edge deployment scenarios where computational resources are constrained.

Key contributions:

  • Designed a lightweight CNN architecture tailored for skeleton-based action recognition
  • Developed a hardware-efficient processor with optimized datapath for real-time inference
  • Achieved real-time performance on embedded devices with minimal resource overhead

Related publications:

  • B. Zhang, J. Han, Z. Huang, J. Yang, X. Zeng, “A Real-Time and Hardware-Efficient Processor for Skeleton-Based Action Recognition with Lightweight Convolutional Neural Network,” IEEE Transactions on Circuits and Systems II: Express Briefs, 2019.
  • J. Yin, J. Han, C. Wang, B. Zhang, X. Zeng, “A Skeleton-Based Action Recognition System for Medical Condition Detection,” IEEE BioCAS, 2019.