CVPR Submission 1925: Christopher Zach, Adrian Penate-Sanchez, Minh-Tri Pham.
"A Dynamic Programming Approach for Fast and Robust Object Pose Recognition from Range Images" (formerly "A Message-Passing Approach for Fast and Robust Object Pose Recognition from Range Images").

The video (supplementary_video.avi, Xvid coded w/o audio) shows several of our own recordings using an ASUS Xtion depth camera.

The top left frame shows the current object to be recognized rendered with color coded object coordindates.
The top right frame shows the best matching feature (object coordinate).
The bottom left frame illustrates the similarity score of the best matching feature.
The bottom right frame renders the current depth map (using normal mapped colors) and the superimposed 3D model according to the best scoring pose estimate.

Feature matching etc. is performed on a 4x subsampled input depth (160x120 resolution). The frame rate is about 5 Hz, and the video is rendered at 10 Hz.
