A CSI-Based Position Independent Gesture Recognition System Using Deep Learning

Wi-Fi-based gesture recognition systems that use CSI data (channel state information) are becoming increasingly popular due to their privacy-preserving and non-intrusive nature. Previous Wi-Fi sensor-based gesture recognition systems typically operated from fixed positions, limiting their usefulness in real-world scenarios. They did not work on all gesture types or position-independent systems. Their system's accuracy was insufficient. To solve these issues, this paper delivers a CSI-based position-independent gesture recognition system using machine and deep learning techniques. This work divided the experimental room into a 3×3 grid and collected training-testing data from different positions. This work used a variety of machine (ML) and deep learning (DL) models to classify four gestures. This paper discovered that the LSTM model outperforms all other methods tested, with an 81.11% accuracy value. The results highlighted that the proposed system improved the accuracy of existing schemes by at least 1.6%.