REAL-TIME HEAD NOD/SHAKE DETECTION FOR CONTINUOUS HUMAN AFFECT RECOGNITION

Head gestures are often used to communicate intent, human affect and to perform conversational functions. Head nod and shake are the most commonly used head gestures in daily life. In this paper we propose a robust system to automatically detect head nod and shake using the Microsoft Kinect in real time. We use Microsoft Kinect SDK to estimate the head pose of the participants. The directions of the head movements determined by head pose is then used as the observation sequence by a discrete Hidden Markov Model (HMM) to determine if head nod or shake occurs. The system achieves a real time recognition accuracy of add number on the test datasets.