Vision and EMG Information Fusion Based on DS Evidence Theory for Gesture Recognition

Aimed at the problem of variety and accuracy of gesture recognition in human-robot interaction, this paper proposes a gesture recognition method based on DS evidence theory for vision and electromyography information fusion. Through the feature analysis of visual images and electromyography signals, human gestures are correctly recognized. The histogram of oriented gradient features of the visual images and the time-domain features of the electromyography signals are extracted respectively to describe the gestures information. Support vector machine is used as classification algorithm. The two description methods are fused at the decision-level by DS evidence theory, which significantly improves the accuracy of gesture recognition. Experiments demonstrate the proposed method has an average accuracy of 93.8% for 36 gestures.KeywordsMultimodal fusionHistogram of oriented gradientElectromyographyDS evidence theory