Pictures are taken from the paper “What and Where: 3D Object Recognition with Accurate Pose” \ I. Gordon, and D. G. Lowe.
The input of the project is a video sweeping a room (only a part of it) and the goal is to add some virtual objects (like furniture) to the frames of the video. We will use structure-from-motion methods to create a simple 3D reconstruction of the room. Then we add the objects and return them back to video (as described in the attached paper). The whole work does not intended to be real-time - we will use Matlab implementation.
In this project we will create a 3D scene and add to it virtual objects.