Improving gesture recognition accuracy with multimodal signals based on fusion of surface EMG and acceleration signals
Shenke Zhang, Wenjie Chen, Xiantao Sun, Cheng Zhang · 2024
Human-computer interaction technology exhibits a broad prospect in the field of rehabilitation assistance, with gesture recognition garnering significant attention as a vital interaction method. The traditional approach to gesture recognition based on surface electromyographic (sEMG) signals encounters challenges such as signal singularity and difficult feature extraction, thereby limiting recognition accuracy. In this paper, we propose a method of combining surface point signals with acceleration signals and additional spatial attention mechanism convolutional neural network (SA-CSAM) to improve the recognition accuracy. Experiments conducted on the NinaPro DB2 dataset demonstrate that our method significantly improves the average accuracy by approximately 19.70%, from 71.96% to 91.66%, when compared to gesture recognition using single sEMG signals. It shows that the SA-CSAM method proposed in this paper is conducive to improving the accuracy of gesture recognition and has potential applications in assisting mobility-impaired elderly people to accomplish daily life tasks.