Accurate 3D Head Pose Estimation with noisy RGBD images

In this paper we propose a novel method for accurate 3D head pose estimation with noisy depth map and higher-resolution color image, which typically produced by popular RGBD cameras such as Kinect. Our method combines the benefit of higher-resolution RGB image and 3D information of depth image. For better accuracy and robustness, features are detected and matched with only color images, the outliers are filtered with depth information. In this way it effectively avoids the influence of depth noise. Several effective outlier filtering rules are introduced. Finally the pose parameters are optimized iteratively using Extended LM (Levenberg-Marquardt) method. To evaluate our method, we built a database of more than 10K RGBD images, with ground-truth poses recorded using motion capture. Both qualitative and quantitative evaluations show that our method produces significant less error than previous methods.