This engineering project presents new video processing and computer vision technique for constructing 3D human hand model from 2D image sequence. The proposed method could be used by computer to understand human hand posture from 2D images. The proposed method incooperate important hand landmarks which can be seen by normal vision and cameras with known simplified human hand model to calculate the angle for each knuckle in the hand. Important landmarks are fingertips webs fingernails and transverse creases. The landmarks are then extracted and matched with the same points on the 3D hand model. The angle for each knuckle are then calculated and the appropriate 3D hand model can be generated. In the experiment, videos of the real human hand were used to test the system. The experiment result shows the generated 3D hand model which poses accordingly to the original hand images. This method works well with index middle ring and little finger. The advantage of this method is that no markers or other equipment are needed therefore the system would need less user cooperation.

Keywords: hand, 3D model, hand posture, hand landmark, 3D reconstruction, monocular view, 2D image sequence, knuckle, index finger, middle finger, ring finger, little finger

The objective of this project is to reconstruct a 3D hand model from 2D image sequence using visual-based approach.