Summary Week4
Kinect Fusion
The paper presents a novel interactive reconstruction system called the Kinect Fusion. It takes live depth data using a moving Kinect camera and then recreates a 3D model of the scene. They also propose a novel GPU pipeline that allows for accurate camera tracking and surface reconstruction in real time. The authors highlight some important features and potential uses of Kinect Fusion. They further describe the implementation of Kinect Fusion and GPU Pipeline. One of the major problems faced during 3D scanning was to make the scene static, but Kinect Fusion overcomes this. Kinect Camera is moved rapidly within a room to reconstruct the 3D model of the scene. The accuracy of Kinect Fusion is also impressive. One of the example in which “DELL” engraving which is hardly 1mm deep is also captured which speaks volumes about the accuracy of depth measurement. Apart from scanning the scene Kinect Fusion also enables scanning smaller physical objects separately once the scene is created. Adding to scanning one can also create virtual objects in the scene which opens a wide array of opportunities of AR applications. Its GPU pipeline consists of 4 main stages.
Depth Conversion, Camera Tracking, Volumetric Integration, Raycasting. The authors have explained the algorithms for implementation of the above mentioned in great detail. Apart from all the above features Kinect Fusion is also interactive. The system makes use property of ICP tracking to keep track of outlier points and by comparing them to the nearest frames any changes in the scene could be detected.
The authors hope to build on the system and make it more memory efficient.
Going out: Robust Model-based Tracking for Outdoor Augmented Reality
The paper discusses “Going Out” a handheld based AR system which generates the environment based on 3D texture modeling instead of GPS using GPS data as a whole. The paper starts off with explaining how using 3D texture models are better than detailed edge models. They also mention the problems incurred on taking the system outdoors. They then go on to explain their system. They start with edge tracking system using 3D building models. They explain how the 3D models are rendered and edge tracking works. The working of the sensors to provide gyroscopic measurement of rotational velocity, 3D acceleration vector, and magnetic field vector to get robust measurement is explained in detail.
The paper also describes implementation of a game on the handheld system to show the performance of the system. They end with publishing the results that show the effectiveness and robustness of the system.
Comments are closed.