Squeeze-Excitation Transformer With Residual Bi-GRU Model for Distributed UWB Based Continuous Gesture Recognition and its Application to Human-UAV Interactions

Attributable to the random features of wireless signal, different environments, user areas, and variabilities in user gestures, wireless gesture recognition becomes more formidable. In this work, a continuous wireless gesture recognition developed by integrating distributed ultrawideband network (DUWBN) and squeeze-excitation transformer with residual bi-gate recurrent unit (SE-T-RB-GRU) model can tackle the above difficulties. It presents distinguished improvements in processing continuous data streams for real-time applications. The details of model training, optimization strategies, and data preprocessing techniques are presented to improve the performance. From the viewpoint of accuracy and training time, the best sequence length from 3 anchors with different heights is achieved. Furthermore, only one subarea including wireless localization is needed for the modeling and the other extended subareas is achieved by coordinate transformationation. A mode filter trigger is also designed to prevent noisy commands. Finally, extensively experimental comparisons with the state-of-the-art methods have average accuracy of 96.31% and an application to human-UAV interactions is implemented. The proposed approach becomes a plug-in module for similar tasks, e.g., a warehouse management system, home appliances.