So I have a bunch of depthmaps (aka range images) of a sculpture, acquired with a handheld Kinect. Now I want to generate a 3-D model of them.
http://pointclouds.org/documentation/tutorials/in_hand_scanner.php#in-hand-scanner (dependent on color; uses ICP) and http://wiki.ros.org/rgbdslam are two ROS-related things that do this kind of thing. http://pointclouds.org/documentation/tutorials/registration_api.php#registration-api has an overview of the process, including an alarmingly large number of options for how to do each step (although apparently SIFT and NARF are the usual feature detectors?), and http://pointclouds.org/documentation/tutorials/interactive_icp.php#interactive-icp has a demo thingy (which uses Blender, but only to generate the input as a .ply file), using the ICP algorithm that is also used in http://pointclouds.org/documentation/tutorials/iterative_closest_point.php#iterative-closest-point. It looks pretty easy to use but I don’t know how well it works for incomplete point clouds with errors.
But I don’t really want a point cloud. I want a triangle mesh so I can render it efficiently and then simulate projecting images onto the sculpture.
NARF, the “normal aligned radial feature”, is https://www.willowgarage.com/sites/default/files/icra2011_3dfeatures.pdf.