Sensor-Based Hand Gesture Recognition Using One-Dimensional Deep Convolutional and Residual Bidirectional Gated Recurrent Unit Neural Network

Hand gesture recognition (HGR) is a crucial domain of study within human-computer interaction (HCI), encompassing applications such as drone operation, virtual and augmented reality, and sign language interpretation. This research introduces an innovative method for HGR utilizing wearable sensors integrated with a sophisticated neural network architecture. We propose an integrated deep residual model called 1D-CNN-ResBiGRU, which amalgamates a one-dimensional convolutional neural network (CNN) with residual bidirectional gated recurrent units (ResBiGRU) to analyze data from wearable sensors to enhance the accuracy and robustness of hand gesture recognition. Our experimental findings indicate that this method attains exceptional accuracy rates of 93.03 and 98.49 $$\%$$ on the GesHome and UWave datasets, respectively, in recognizing diverse hand motions. The achievement substantially surpasses current methodologies, exhibiting enhancements of up to 7.16 percentage points compared to state-of-the-art models. The suggested system demonstrates notable proficiency in managing intricate motions and ensuring uniformity across user demographics. It is appropriate for utilization in human-computer interaction, smart home management, and assistive technologies. Ablation experiments further confirm the efficacy of each component in our design, with the 1D-CNN blocks and ResBiGRU component greatly enhancing the model’s capability to accomplish.