Summary

The Kinect opens new possibilities regarding the fast acquisition of 3-D models of objects. Using a single Kinect image only, it is already possible to reconstruct a good approximate model of an object lying on a table. If more precision is required, multiple Kinect views can be merged using fiducials markers to estimate the pose of the camera in each image, and a complete mesh can be reconstructed after proper alignment. The 3-D models acquired by Kinect can be further recognized in new scenes thanks to powerful 3-D descriptors that allow you to robustly recognize point clouds, even with small viewpoint changes.