Departmental Bulletin Paper Diminished Reality based on Kinect Fusion

Li, Qiaozhi

While Augmented Reality allows adding virtual objects over a real scene, Diminished Reality can remove real objects from a scene. Currently existing multi-view based approaches and image completion based approaches to DR with 2D sensors. In this paper, we propose a practical approach to diminished reality for indoor environment with 3D model creation based on Kinect 3D sensors. Kinect can capture RGB information and per-pixel depth information of scenes at same time, and it can use the depth information to create 3D model. Our approach is based on the 3D model to search the region of the objects which will be removed. First we get the RGB and depth information of a scene from Kinect and get 3D model based on Kinect Fusion. After that an object is added into the scene. We compare the scene with original scene and process surface images acquired from Kinect fusion to find the accurate region of the object. At last we replace the data and diminish the object in real time. The solution uses Kinect fusion technique to create models and OpenCV for image processing. Our approach achieves a good result that we can diminish the objects clearly and restore the scene. Our approach provides real-time Diminished Reality for indoor environment.

Number of accesses :  

Other information