Gesture recognition using a depth camera for human robot collaboration on assembly line

We present a framework and preliminary experimental results for technical gestures recognition using a RGB-D camera. We have studied a collaborative task between a robot and an operator: the assembly of a motor hoses. The goal is to enable the robot to understand which task has just been executed by a human operator in order to anticipate on his actions, to adapt his speed and react properly if an unusual event occurs. The depth camera is placed above the operator, to minimize the possible occlusion on an assembly line, and we track the head and the hands of the operator using the geodesic distance between the head and the pixels of his torso. To describe his movements we used the shape of the shortest routes joining the head and the hands. We then used a discreet HMM to learn and recognize five gestures performed during the motor hoses assembly. By using gesture from the same operator for the learning and the recognition, we reach a good recognition rate of 93%. These results are encouraging and ongoing work will lead us to experiment our set up on a larger pool of operators and recognize the gesture in real time. (C) 2015 The Authors. Published by Elsevier B.V.