Abstract

This research investigates the development of predictive models for recognizing dual-hand gestures using surface electromyography (sEMG) signals. The main focus was put on dual-hand gestures, an area that remains a gap in research and is essential to advance technologies in human-computer interaction. Most of the previous systems are designed for single-hand gestures, leading to low accuracy for more complex dual-hand gestures. The basic objective of this research is to develop a robust predictive model which enhances the recognition of dual-hand gestures significantly through analysis of the NinaPro DB1 database. We used two approaches to explore into the problem: one using pre-trained models and the other with our ensemble learning method. The second approach is a novel hybrid model that consists of a Convolutional Neural Network (CNN) and a Gated Recurrent Unit (GRU). In that, the spatial and temporal features from the data communicated by the sEMG signals were captured for the description of the complex dynamics of dual hand gestures. The application of the models provided different levels of success. The pre-trained model of VGG16 resulted in an accuracy of 60%, illustrating the complexity of adapting sEMG signals for image-based neural networks. The hybrid CNN-GRU model yielded better accuracy, with a first set of gestures achieving 83% and a second set achieving 85% over the dataset. These two low-level models were combined using a higher model that utilizes random forest, which can be used through action mapping to support various operations. The higher model achieved an impressive 99% accuracy, indicating the success of combining CNN and GRU for this type of data. The high classification performance of the hybrid model infers success in effectively handling the spatial-temporal complexities of dual-hand gestures.

Statistics