This work proposes an accurate marker-less fingertip detection method under single camera. The moving skin regions are analyzed via enhanced mixture models. The models with adaptive learning rates can separate the backgrounds and moving skins effectively. For multiple fingertip detection, we propose an algorithm based on the likelihood computation of contour and curvature information. Furthermore, finger width validation and error correction via temporal information is used to improve the detection accuracy. The experiments have shown that the proposed method is robust and flexible. Finally, we implement a human computer interface system to test the effectiveness of the proposed framework.