Gesture recognition using elastic data glove based on FFT-BIGRU

Gesture recognition, as one of the important components of human-computer interaction, has a wide range of applications in important fields such as rehabilitation medicine, machine control, and virtual reality. Gesture recognition based on wearable data gloves has attracted widespread attention due to its strong universality and scalability. This paper utilizes an elastic data glove independently developed by our team to collect data. The glove integrates independently developed highprecision stretchable angle sensors and common inertial sensors, which can effectively recognize the bending and spatial position of the fingers. The innovative use of angular velocity thresholds for gesture segmentation, the utilization of linear interpolation and combination mean for data resampling, and the development of a network structure combining Fast Fourier Transform (FFT) with Bidirectional Gated Recurrent Unit (BIGRU) not only effectively captures the periodic patterns in the data but also handles long sequence dependencies. Achieving an accuracy of 94.70% in the classification of 14 sign language gestures, this approach outperforms other classical models. Ablation experiments have validated the effectiveness of the proposed network modules and stretchable angle sensors.