RFID Gesture Recognition Based on Convolutional Neural Networks

Gesture recognition technology has gradually become a key tool in HCI (human-computer interaction), widely applied in virtual reality, smart homes, sign language translation, and other fields. Although traditional contact-based gesture recognition methods perform well in terms of accuracy, they pose significant inconveniences in practical use. In contrast, non-contact methods effectively overcome these issues. In particular, RFID (Radio Frequency Identification)-based gesture recognition technology, with advantages such as non-contact operation and privacy protection, has emerged as a promising solution. To this end, this paper proposes an innovative gesture recognition method that converts RFID signals into images and employs 2D-CNN (convolutional neural network) for image classification, effectively addressing the limitations of traditional methods in spatiotemporal feature extraction. Experimental results demonstrate that with only a four-layer CNN, the model achieves an accuracy of 98.33% on the test set and exhibits excellent real-time performance, meeting the requirements for lightweight applications.