| With the fast development of computer vision technology in recent years, face detection and face tracking have been studied extensively, such as face recognition, intelligent human-computer interaction, video conference and so on. At the same time, robot technology has developed rapidly, and human robot interaction has become an important research area.Most robots are working in embedded environment because of its advantages: cost-effective, low-power and so on. With the rapid development of embedded hardware and software technologies, computing real-time video in embedded environment becomes a possibility. The major work of this paper is as follows:We present a face detection method based on image sequences. Our method mainly has improved in two aspects. First, we use motion and skin detection to obtain foreground. Second, the method of improved-AdaBoost is used to detect face in foreground. So we achieve low time complexity.We also present a method of real-time face tracking. There are two improvements as below. First, pyramid Lucas-Kanade optical flow algorithm with distance limitation is used to track face region, and the complexity of the algorithm is reduced quickly by foreground extraction. Second, when facial features are lost seriously, our method will re-initialize these features. Finally, we apply these methods in our robot platform to obtain good detecting and tracking performance. |
