A Dual-Stream Fusion CNN-LSTM Gesture Recognition Algorithm for Command and Control Systems
Shengjie Han, Guang Li, Xiaotong Ding, Meng Lu · 2023
In this study, we propose a novel gesture recognition algorithm based on dual-stream feature fusion CNN-LSTM, aimed at addressing the limitations and deficiencies of existing gesture recognition methods in human-machine interaction systems within the military command and control sphere. We employed Blackman windowing along with wavelet thresholding and dynamic zero-padding algorithms to suppress clutter and enhance data robustness. Utilizing the Multiple Signal Classification technique, we computed the angular parameters, which were synergized with distance and velocity parameters to create range-Doppler matrix plots and angle-time graphs. These feature maps were fed into a dual-stream CNN-LSTM network for feature extraction and classification, thereby achieving high-precision gesture recognition. Notably, this advanced system achieves a stable and precise gesture recognition accuracy rate of 98.25%, offering an efficient and secure avenue for enhanced human-machine interaction in the military domain.