Fig. 1
From: Robust head pose estimation based on key frames for human-machine interaction
Pipeline of the proposed head pose estimation method. Top: offline framework. The input is a RGB-D stream of the person to track. Rough pose estimation is done (red block), and most relevant frames are selected (yellow block). Then, an optimization process improves the pose accuracy (green block). Bottom: online pose estimation process. The current image is shown at left. The key frame with the best matching score is selected and depicted at right. Finally, ICP refines the estimation (right image)