Research Article

Optimization and Soft Constraints for Human Shape and Pose Estimation Based on a 3D Morphable Model

Figure 9

The local estimation Result. When the motion is slow, the estimation result using local optimization method is correct (frame no. 1–60), Once it changes faster, the tracking will make error no. 70 frame, this will need global optimization to fix. Using annealed particle filtering, we need 100 particles and 15 iterations, after we add the pose prior constraint for the particles, we just need 25 particles and 10 iterations.
715808.fig.009