Accelerometer-based Hand Gesture Recognition for Human-Robot Interaction

This paper presents a computational solution towards human-robot interaction using a wrist-mounted tri-axial accelerometer. This is tackled as a three-fold gesture recognition problem with a gesture set including six different gestures, namely right, front, left, back, up and circle. Given the sparsity of gestures, an adaptive segmentation technique is employed as a means of spotting potential segments of interest within the signal. Relevant features are a posteriori calculated from the spotted segments. Ultimately, a range of five state-of-the-art classifiers are employed for the classification of the gestures. The results achieved, with an average classification accuracy of 95.85%, show a great contribution towards modern human-robot interaction technologies.