A real-time head nod and shake detector using HMMs

摘要

Among head gestures, nodding and shaking are commonly used to communicate intent, emotion and to perform conversational functions. In this paper, a new method is proposed to detect head-nodding and shaking in real time from video streams. The system is composed of eye tracking and head motion decision. The eye tracking is finished in two steps: face detection and eye location. Face detection obtains the face region using the cascaded classifier constructed by AdaBoost algorithm. Eye location is performed only in the detected face region. The changes of the eye's x-position and y-position indicate the direction of head movement. Then these directions are used as observations by discrete Hidden Markov Models (HMMs) to detect whether a head nod or shake occurs. The system is trained and tested on the dataset collected from laboratory members, and achieves a recognition accuracy of 85%. An interesting opinion survey program is constructed using the algorithm.