Gesture Recognition Using Hand Region Estimation in Robot Manipulation

In this study, we propose a gesture recognition method for operating an unmanned aerial vehicle (UAV) equipped with only a single camera. This method uses a neural network to estimate the 2D coordinates of a hand region in an image from color information. From the time series data of hand coordinates, we estimate the gesture motion by 1D Fast Fourier Transform. The proposed method identifies seven types of gestures and manipulates the UAV. The accuracy and usefulness of this method are confirmed by experiments of UAV movements in response to gestures.