Welcome to Easy Open3D, the easy-to-use wrapper around (as well as utility functions and scripts for) some of Open3D ‘s registration functionality.

What is registration?

In 3D data analysis, the term registration usually refers to the process of aligning two partly overlapping point clouds such that the result is a merged version of the input. It is frequently used to obtain a 3D scan of a large scene by stitching together several smaller parts. In robotics however, we can use the same algorithms to register a 3D model of an object (called source) we want to manipulate to the part of the 3D scan (obtained through depth sensors; called target) of the scene the robot operates in to find its pose in this scene. The pose of an object consists of its rotation and translation in some coordinate frame, typically the robots head frame, camera frame or world frame.

A simple (yet slightly contrived) example

Say we want to find the pose of Blender mascot Suzanne (the blue ape head on the chair) in this scene:

We are also given a 3D model of her (the source) and a depth image of the scene (the target):