Studies on hand motion recognition based on biosignals have become popular as such recognition can be applied to various input interfaces and motion measurements for human–robot/computer interaction. In recent years, many machine-learning-based technologies have been developed to analyze such biosignals more accurately. Among various possible biosignals, we focus on forearm deformation which is an alternative source of information for hand motion recognition. The activities of surface and deep layer muscles, tendons, and bones can be extracted from forearm deformation in a non-invasive manner. In this study, a hand motion recognition system is proposed based on forearm deformation. By using machine-learning-based technology, the proposed method can be applied to various users and various measurement conditions. First, a distance sensor array is developed to measure forearm deformation. Then, we test and verify the suitableness of three types of machine-learning-based classifiers (k-NN, SVM, and DNN) using the measured forearm deformation. In experiments, we verified the accuracy of the proposed system with various users. We also test the system for different elbow postures, and when measuring the data over the clothing.
展开▼
