Investigates Image data fusion techniques that combine image and track data from multiple sensors to achieve improved accuracies and more specific inferences than could be achieved by the use of a single sensor alone. Our aim is to explore the state-of-the-art image processing algorithms for achieving effective data fusion as in:
3D Head Tracking & Angle Detection
1. Introduction
A. 6DOF head tracking with single webcam is challengeable for head gesture interface. So modeling 3D cylinder on head, it is used for solving the LK method.
2. Framework of Algorithm
A. Obtain the Face region by Adaboost face detection
B. Modeling 3D cylinder model & assign the 2D template
C. Estimate the motion parameter using the LK method
D. Update 3D cylinder model & 2D template
E. Calculate the 3D face angle
F. Repeat C. ~ E.
3. Details
A. Parameter estimation by LK method
i. Motion parameter
ii. Minimize Error function
Solving the minimizing problem by Lucas-Kanade method
B. Guide by 2d tracker
i. If translation is large, then it would be fail to track. So preventing this case, global motion is estimated by multi-cue mean-shift algorithm
ii. Block diagram of Multi-cue mean-shift algorithm
4. Demo
