Table 1.
Summary of multi-stream techniques utilized in human action recognition.
Table 2.
Summary of multi-branch techniques for human action recognition.
Fig 1.
The proposed methodology.
Fig 2.
Architecture of the enhanced multi-stream adaptive graph convolutional network (EMS-AGCN).
Fig 3.
Architecture of the multi-branch adaptive graph convolution network (MB-AGCN).
Fig 4.
Architecture of the proposed adaptive fusion model.
Fig 5.
(a) Illustrates the structure of the spatio-temporal block within the main branch, whereas (b) Provides a detailed description of the adaptive graph convolutional layer.
Fig 6.
Performance impact of the proposed features on the baseline model.
Table 3.
Comparative analysis of accuracy across various input modalities on the NTU-RGBD 60 dataset.
Table 4.
Comparisons of the accuracy of multi-stream with different input modalities on the NTU-RGB+D 60 dataset.
Fig 7.
Cross-view performance evaluation of EMS-AGCN on the NTU-RGB+D 60 dataset.
Table 5.
Comparisons of the accuracy with different input modalities on the NTU-RGBD 60 dataset.
Table 6.
Comparisons of the test results with state-of-the-art methods on the NTU-RGB+D 60 dataset.
Fig 8.
Comparison of EMS-AGCN and MB-AGCN.
Fig 9.
Accuracy of MB-AGCN using joint and bone modalities.
Fig 10.
The confusion matrix of the MB-AGCN model applied to the cross-view protocol of the NTU-RGB+D 60 dataset.