Description

When scanning in 3D, each single scan results in a depth image, which has to be merged to the other scans as accurately as possible to build the surface. Iterative Closest Point (ICP) is a standard algorithm for the registration of range scan data, with many variants. But it needs a good guess, which often must be provided or corrected by the user, in order to converge. Since this is a big problem in practice, we use techniques from shape matching, able to detect partial similarity, in order to provide that initial guess.

Task

Use a suitable rigid and partial shape matching technique, e.g. Bag of Words, to find the correspondences between two range images. The best correspondences translate into their affine transform, registering them with each other.

Requirements

Good programming skills. Interest in algorithms.

Environment

C/C++, platform-independent

• Computer Graphics Group
• Courses
• Projekte
• 3D Scan Registration