Gesture-Based Human-Robot Interaction

The paper proposes a method for human robot interaction through gestures. Even if there are different methods that perform gesture recognition, many of them are tested in ideal conditions. This paper proposes a convolutional neural network that is used for gesture recognition applied for human-robot interaction in different environmental conditions: such as: blurring, low or high illumination, low quality image or losing part of frames. We started from a pre-trained ResNeXt 101 model and we increased the accuracy by adding a preprocessing stage for improving the quality of image, applying data augmentation, using different optimisers and learning rate scheduler. The proposed model obtained an improvement of 0.9% compared to the baseline model. Online evaluation of gesture recognition was performed using the Turtlebot robot.