3D Facial Expression Acquisition And Recognition Based On Kinect

At present, face tracking and facial expression recognition in human-computer interaction, image retrieval, intelligent monitoring, video conference, virtual reality, medical and automotive electronics and other fields has a broad prospect of application and its implementation method research has important theoretical significance and engineering application value, therefore the more and more attention of experts and scholars.This paper use Kinect sensor as the data acquisition device for 3d data expression recognition research. In this study, we use Kinect to collect depth and color data, facial triangular mesh data obtained conversion. Use the head bone is proposed for facial method for rapid detection and tracking, and to detect the facial 3d data of calibration normalization processing and PCS. After using Kernel Principal Component Analysis was carried out on the facial feature information extraction, and combining the Facial Action Coding System design of the Support Vector Machine classifier analysis of extracting the facial features of information classification, identify the corresponding state of emotion.According to the results of this study was designed and implemented after facial expression recognition system, and invited five classmates in turn made six of the most common expressions(happy, surprise, fear, sadness, disgust and anger), which can identify 360 times recognition experiments were conducted in three group, make two contrast is formed three groups of experiments to test the performance of facial expression recognition system, and verify the effectiveness of the proposed research method.Three groups of experiments the final average recognition rate of 68.3%, 67.5% and 51.7% respectively, compared with the existing research recognition rate is low, it is also the problems existing in the research. But this system is able to have the face to modify(with glasses) or head posture not ideal(the face and the camera is opposite direction into 30 degree Angle) still maintained the recognition rate, overcome the general expression recognition for illumination, head posture and shade condition such as required.