Week 5 Summaries
KinectFusion: Real-time 3D Reconstruction and Interaction Using a Moving Depth Camera
Kinect is a motion sensing input device, creating real-time depth maps containing discrete range measurements of the physical scene. The advantage of using Kinect is the quality of depth sensing in real time with low cost. However, the data are noisy and contain numerous holes. KinectFusion is developed to use a moving Kinect so as to create a single high-quality and accurate 3D model.
KinectFusion enables AR systems to map the world in real time. Raytraced rendering effects can be calculated to create shadows and lighting. The challenging problem in this case is the prolonged interaction with the scene. A GPU-based pipeline is extended to approximate the camera motion. This locks the tracking camera onto the background and ignores the foreground user for camera pose prediction.
Different algorithms presented in this study to do real time 3D tracking, reconstruction, segmentation, rendering and interaction. These techniques demonstrated compelling results in stabilizing, tracking, and also improving reconstruction quality. In addition, foreground objects are segmented separately from the background.
A Motion-Stabilized Outdoor Augmented reality system
Localization is one of the problems in outdoor augmented reality. Traditionally, different techniques and sensors were used to get the accurate location of the device such as using GPS, magnetic compass, and inertia sensors. These tools and techniques may not be available anywhere and they are prone to errors and drifts.
This study demonstrate a model based hybrid tracking system for outdoor AR. The system combines the edge-based tracker method with different sensor data to get the accurate position. The edge tracking system relies on the 3D model of the scene, generating the model in every frame. For each frame, a set of corner-like features are extracted using FAST corner algorithm.
The system was tested in two different sites against accuracy, robustness, performance, and dynamic behavior. The results show a systematic error, illustrating a closer appearance of the buildings. Overall, the system was efficient to be used but it had some limitations.
Comments are closed.