A hybrid approach for static hand gesture recognition with integrated BiGRU-BiLSTM and sequential self-attention mechanism
摘要
This paper proposes a deep network named BiGrLNet for static hand gesture recognition. It consists of a multibranched convolutional module and an attentive network module integrated with BiGRU-BiLSTM and sequential self-attention mechanisms. The multibranched convolutional module, designed with varying dilation rates, enhances the ability of the system to capture multiscale spatial features essential for distinguishing intricate details of hand gestures. Sequential self-attention mechanisms are employed at multiple stages to allow the model to focus on the most relevant parts of the input features, improving representation and recognition accuracy. The BiGRU unit processes the features to capture dependencies in both forward and backward directions, ensuring robust learning of spatial patterns. Subsequently, the BiLSTM unit further refines these dependencies, leveraging its capacity to maintain long-term contextual information. The performance of the proposed model is investigated based on accuracy, precision, recall, and the F1-Score on ten benchmark datasets: ASL Alphabets (A