Opto-Electronic Engineering, Volume. 51, Issue 6, 240072-1(2024)

Design and implementation of edge-based human action recognition algorithm based on ascend processor

Dongdong Zhao, Liang Lai, Peng Chen*, Hongchao Zhou, Yiran Li, and Ronghua Liang
Author Affiliations
  • College of Computer Science and Technology, Zhejiang University of Technology, Hangzhou, Zhejiang 310023, China
  • show less
    Figures & Tables(13)
    Overall framework of the YOP-Ascend-STGCN algorithm
    Skeleton explicit and implicit association structure diagram of the running action
    Spatial-temporal modeling diagram of the skeleton information
    Explicit and implicit subgraph partitioning methods
    Ascend-STGCN network structure diagram
    System hardware structure
    System software process
    Histogram of the comparison data for each action
    Screenshots of inference results for each action
    Device structure diagram
    Screenshot of inference results for simulated actions
    • Table 1. Comparative experiments on the KTH dataset

      View table
      View in Article

      Table 1. Comparative experiments on the KTH dataset

      AlgorithmTop-1/%Params/MFlops/G
      ST-GCN79.173.685.88
      AS-GCN83.087.0010.69
      ST-TR83.8311.5113.88
      PoseConv3D85.673.2316.1
      This paper84.173.014.57
    • Table 2. Ablation experiments on the KTH dataset

      View table
      View in Article

      Table 2. Ablation experiments on the KTH dataset

      AlgorithmTop-1/%Params/MFLOPs/GEpochs/轮
      ST-GCN (EC+TC)79.173.685.8850
      EC+TC+AE81.672.934.5750
      EC+TC+IC82.503.795.8850
      EC+TC+SA83.333.685.8850
      Ours (EC+IC+TC+AE+SA)84.173.014.5750
    Tools

    Get Citation

    Copy Citation Text

    Dongdong Zhao, Liang Lai, Peng Chen, Hongchao Zhou, Yiran Li, Ronghua Liang. Design and implementation of edge-based human action recognition algorithm based on ascend processor[J]. Opto-Electronic Engineering, 2024, 51(6): 240072-1

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category:

    Received: Mar. 25, 2024

    Accepted: May. 23, 2024

    Published Online: Oct. 21, 2024

    The Author Email: Peng Chen (陈朋)

    DOI:10.12086/oee.2024.240072

    Topics