Tracking texture-less, shiny objects with descriptor fields

Our demo demonstrates the method we published at CVPR this year [3] for tracking specular and poorly textured objects. Instead of detecting and matching local features, we retrieve the pose in the input images by aligning them with a reference image exploiting dense optimization techniques. Our main contribution is an efficient novel local descriptor that can be used in place of the intensities to make the alignment much more robust. Our approach, which requires only a standard, monocular camera (no need for a depth sensor), is of great interest for all Augmented Reality applications involving shiny, texture-less objects, such as those typically encountered in industrial environments.