Hand Tracking

One of the objectives of this project is real time tracking and registration of the human hand, and fingers in 3D space. This information will be used to manipulate objects in a virtual environment; therefore the desired precision of the estimates is much higher than that of most of the existing studies, which mainly focus on the communicative use of the hand and utilize the estimates to recognize gestures.

The input to the system will be synchronized image sequences of the moving hand from different view points (4-8). The output for each frame of the input sequences will be the position and the orientation of the palm in 3D space and the joint angles of the fingers. The system will have the following features:

	At the start-up the hand will be assumed to be roughly at a predetermined initial state.
	The hand will be modeled using a 3D cylinder hierarchy complemented with static and dynamic constraints of hand motion.
	The hand model will be calibrated for each user through a simple calibration procedure.
	In VGX the user is required to wear gloves. Therefore some markers with different colors will be placed on the glove to increase robustness in feature extraction.
	A silhouette based 3D reconstruction of the hand will be performed and the state will be estimated by deforming the 3D hand model to fit to the reconstructed hand and extracted features.
	In tracking Kalman filter will be utilized to predict the next state of the hand.