Table 1 Summary of literature review.
Study | Algorithm(s) used | Problem addressed | Key features | Limitations |
|---|---|---|---|---|
CNNSa-LSTM | Spatial–temporal gesture recognition | Attention-based CNN, LSTM for sequences | Risk of overfitting on small datasets | |
AEGWO-Net (Autoencoder + GWO) | Dimensionality reduction and feature optimization in ASLR | High accuracy, adaptive GWO parameters, validated on 6 datasets | Still face challenges in real-time applications | |
Lightweight CNN | Real-time gesture recognition | Low computational cost, embedded deployment | Limited temporal modeling | |
Dual-view Transformer | Occlusion-aware SLR with multi-view inputs | Pose integration, egocentric data capture | Needs extensive training data | |
Ego-SLD (CNN-LSTM) | Egocentric dynamic SLR | Realistic dataset, wearable capture | Requires synchronization and noise control | |
Skeleton + Self-attention | Manual and non-manual gesture modeling | High accuracy on AUTSL, multi-modal | Complex model integration | |
GWO for CNN tuning | CNN architecture optimization | Faster training, improved generalization | Parameter sensitivity |
