Research On Face Recognition For Human-Robot Interaction

For the widely use of mobile-robot in service area, people pay more and more attention to the friendly method of human-robot interaction, like face recognition. This thesis researches the core technologies of face recognition, such as the technology of face location, face feature recognition and tracking, which can effectively operate on mobile-robot in real time in various illumination and facial gesture.The thesis summarizes the research and development of human-robot interaction and face recognition. In order to track the rotation of face center, skin color detection and image difference are presented to locate face under complicated background, bilateral symmetry of gradient direction and geometric relationship of face features are used to extract midline of face and position of face features, then the center of eyes is obtained, which nearly represents the center of the face for tracking. This method can solve the tracking problem due to face features' shape change caused by face rotation in certain angle range.To solve the problem of the tracking of facial gesture, skin color detection and ellipse approximation are used to locate face; gradient projection of special area is applied to extract the features in initial frame, then Kalman filter is adopted to forecast the position of features in followed frames which is renewed by gradient projection in small feature windows. Taking the average position of features as face center for tracking, the method realizes the comprehension of simple facial gesture.This method can track face features in both horizontal and vertical rotation angle, it can be used for any kind of human and in more condition compared to template match. Experimental results and further experimental data analysis show the method's validity and practicability.