Comparison of gesture classification methods with contact and non-contact sensors for human-computer interaction

Abstract

Classification of signals that are received from the human body and control systems is one of the most important subjects of the machine learning application. In this study, classification algorithms were used to classify electromyography and depth sensor data. First, electromyography and joint angle data were obtained from software developed in Python environment. Five different types of movements have been identified for classification and thousand different samples have been collected as training for each of these movements. Support Vector Machine, Random Forest, and K-Nearest Neighbour algorithms were used for classification. To measure success algorithms, results have been compared for achieving criteria. The results show which of three different algorithms was the most successful on two different sensors. While Random Forest provides the best results for non-contact sensor, K-Nearest Neighbour produces the best results for contact sensors. This paper evaluated the classification success of two different sensors. The results will be utilized in online classification to control a graphical user interface.