Unlocking Seamless Gesture-Based Interfaces: Hand Gesture Recognition with 3DCNN and LSTM

In the symphony of human communication, the fusion of words and gestures creates a nuanced expression that transcends mere verbal exchange. Replicating this intricate dance in the realm of Artificial Intelligence (AI) has become a compelling challenge, with innovations aiming to make digital experiences more human-centric. This work addresses the challenge of gesture recognition by leveraging 3D Convolutional Neural Networks (3DCNN) and Long Short-Term Memory (LSTM) to analyze hand gestures, focusing on time-series data derived from human pose detection. While both 3DCNN and LSTM have been used separately in related fields, their combination for gesture recognition has not been widely explored. The proposed model effectively extracts spatial features using 3DCNN and captures temporal dynamics with LSTM, leading to a 94% accuracy in ten-class classification using a five-fold cross-validation on 400 videos from four participants. Consequently, this work potentially marks a transformative era in gesture recognition, enabling contactless operation of hardware devices for more interactive experiences.