Week 5 Summaries
KinectFusion: Real-time 3D reconstruction and Interaction Using a Moving Depth Camera
This paper talks about a real-time 3D reconstruction and interaction system called KinectFusion, which is built based on Kinect. This system provides 3D scene reconstruction as well as interaction in real-time. 3D models in the scene are reconstructed from the data captured by Kinect depth camera and textures of 3D model are generated from Kinect RGB data.
For the 3D model reconstruction, the system firstly convert depth map into 3D vertices and normal in the coordinate space of camera. Then the relative transform of Kinect post is computed by using Iterative Closest Point algorithm. The global pose of Kinect is computed by incrementally applied the relative transforms of former poses. After getting the global pose of Kinect, the 3D vertices are converted into global coordinate and are used for surface reconstruction using Volumetric Integration method. Finally, the volume is raycast to extract view of the implicit surface and rendered to the user. Most of these processes are done by GPU to achieve the real-time goals.
This system also simulates real-world physics in the virtual scene, which provides several interactive ways for AR applications. The system can also detect user’s touching on arbitrary surface. Overall, I think this system has great potential for creating interesting indoor AR applications.
Going out: Robust Model-based Tracking for Outdoor Augmented Reality
This paper introduces a model-based hybrid tracking system used in out-door urban environment. In out-door urban environment, a computer vision-based tracking system can usually provide better result than GPS because GPS is not robust against shadowing from buildings and signal reflections. This tracking system combined several tracking approaches to provide robust result. An edge-based tracker which is based on 3D texture model is used to provide accurate localization. They also used an inertial sensor and magnetic field measurements to give more accuracy result.
The edge-based tracking system tracks the pose of camera by tracking the motion of the camera from the prior pose. The system first rendered a view of the textured 3D city model from the prior camera pose. Then edges are extracted using Canny edge detector and projected back onto the model to get 3D coordinates. The posterior pose is then yield by updating with measurements. Besides the edge-based tracking system, inertial measurements help to make the system more robust for fast motions. Different measurements are fused by using an extended Kalman filter. They also add a recovery component to this system to handle transient occlusions. At last, this system was evaluated by four aspects: accuracy, robustness, robustness and dynamic behavior.
Comments are closed.