A5114488641- Advanced Vision and Imaging
- Computer Graphics and Visualization Techniques
- Scheduling and Optimization Algorithms
- Advanced Manufacturing and Logistics Optimization
- Advanced Image and Video Retrieval Techniques
- Wireless Communication Networks Research
- 3D Shape Modeling and Analysis
- Optical measurement and interference techniques
- Data Visualization and Analytics
- Image Processing Techniques and Applications
- Human Motion and Animation
- Image and Video Stabilization
- Advanced Wireless Network Optimization
- Interactive and Immersive Displays
- Simulation and Modeling Applications
- Visual Attention and Saliency Detection
- Assembly Line Balancing Optimization
- Advanced Wireless Communication Techniques
- Image and Video Quality Assessment
- Video Analysis and Summarization
- Cultural and Historical Studies
- Indoor and Outdoor Localization Technologies
- Manufacturing Process and Optimization
- Human Pose and Action Recognition
- Digital and Traditional Archives Management
Korea Advanced Institute of Science and Technology
2009-2024
Kai Research (United States)
2016-2017
Korea University
1996-2002
This paper presents Rich360, a novel system for creating and viewing 360° panoramic video obtained from multiple cameras placed on structured rig. Rich360 provides an as-rich-as-possible experience by effectively resolving two issues that occur in the existing pipeline. First, deformable spherical projection surface is utilized to minimize parallax cameras. The deformed spatio-temporally according depth constraints estimated overlapping regions. enables fast efficient parallax-free stitching...
We propose a real-time convolutional neural network (CNN) training and compression method for delivering high-quality live video even in poor environment. The server delivers low-resolution segment along with the corresponding CNN super resolution (SR), after which client applies to order recover high-resolution frames. To generate trained real-time, our rapidly increases accuracy by promoting overfitting property of while also using curriculum-based training. In addition, assuming that...
High-quality depth painting for each object in a scene is challenging task 2D to 3D stereo conversion. One way accurately estimate the varying within an image utilize existing models. Automatic pose estimation approaches based on 2D-3D feature correspondences have been proposed obtain from given model. However, when model not identical target object, previous methods often produce erroneous vicinity of silhouette object. This paper introduces novel model-based method that effectively...
We present a hybrid approach that segments an object by using both color and depth information obtained from views captured low-cost RGBD camera sparsely-located cameras. Our system begins with generating dense of each target image Structure Motion Joint Bilateral Upsampling. formulate the multi-view segmentation as Markov Random Field energy optimization on graph constructed superpixels. To ensure inter-view consistency results between images have too few features, our local mapping method...
This paper introduces ScreenX, which is a novel movie viewing platform that enables ordinary theatres to become multi-projection theatres. the general public enjoy immersive experiences. The left and right side walls are used form surrounding screens. display environment delivers strong sense of immersion in viewing. However, naïve content on results appearance distorted images according location viewer. In addition, different dimensions width, height, depth among may lead Therefore, for...
We present a new interactive playback method to enhance 360° viewing experiences. Our automatically rotates the virtual camera of panoramic video (360° video) player during guide viewer through most important regions video. With this method, can watch with minimum efforts find events in scene both (e.g., HMD) and less-interactive PC TV) environments. To estimate importance each direction, we combine spatial temporal saliency cluster-based weighting. A maximum backward cumulative volume...
Abstract Accurate depth estimation is a challenging, yet essential step in the conversion of 2D image sequence to 3D stereo sequence. We present novel approach construct temporally coherent map for each The quality estimated high enough purpose of2D conversion. Our first combines video into panoramic image. A user can scribble on this single specify information. then propagated remainder This remapped original and used as initial guess individual greatly simplifies required interaction...
Abstract Realistic character animation requires elaborate rigging built on top of high quality 3D models. Sophisticated anatomically based rigs are often the choice visual effect studios where life‐like CG characters is primary objective. However, a with muscular‐skeletal system very involving and time‐consuming process, even for professionals. Although, there have been recent research efforts to automate either all or some parts complexity nonetheless opens up new challenges. We propose...
The needs for a more immersive viewing experience in modern cinemas have given birth to novel movie theater technologies such as Barco's Escape and CJ-CGV's ScreenX; which utilize the side walls of project peripheral content auxiliary main projected center screen. However, because theaters that support application are each different structure, producers need modify is cumbersome task. In addition, recent advances Virtual Reality technology allows users create unexplored solutions area games,...
Abstract Computer graphics is one of the most efficient ways to create a stereoscopic image. The process CG generation is, however, still very inefficient compared that monoscopic generation. Despite stereo images are similar each other, they rendered and manipulated independently. Additional requirements for disparity control specific lead even greater inefficiency. This paper proposes method reduce inefficiency accompanied in creation system automatically generates an optimized single...