Gesture ToolBox: Touchless Human-Machine Interface Using Deep Learning

Human-Computer Interaction (HMI) is useful in sterile environments such as operating rooms (OR) where surgeons need to interact with images from scanners of organs on screens. Contamination issues may happen if the surgeon must touch a keyboard or the mouse. In order to reduce contamination and improve the interactions with the images without asking another team member, the Gesture ToolBox project, based on previous methods of Altran Research, has been proposed. Ten different signs from the LSF (French Sign Language) have been chosen as a way to interact with the images. In order to detect the signs, deep learning methods have been programmed using a pre-trained Convolutional Neural Network (VGG-16). A Kinect is used to detect the positions of the hand and classify gestures. The system allows the user to select, move, zoom in, or zoom out images from organs on the screen according to the recognised sign. Results with 11 subjects are used demonstrate this system in the laboratory. Future work will include tests in real situations in an operating room to obtain feedback from surgeons to improving the system.