|
Image
Perspectives
|
Project description with references Exact
Inference in Multi-label Higher order CRFs for
Single View 3D Reconstruction Randomized
Trees for Human Pose Detection Learning
priors for calibrating families of stereo cameras A Factorization based Self-Calibration
for Radially Symmetric Cameras |
|
|
Theory and Calibration Algorithms for Axial Cameras ( ACCV’06 ) |
|
|
Generic Self-Calibration of Central Cameras ( Omnivis’05 ) |
|
|
Towards Complete Generic Camera Calibration ( CVPR’05, INRIA Report 5562, ECCV’04, INRIA Report 5058 )
We consider the problem of completely calibrating a highly generic imaging model that consists of a non-parametric association of a projection ray in 3D to every pixel in an image. We describe a complete calibration approach that should in principle be able to handle any camera that can be described by the generic imaging model. Initial calibration is performed using multiple images of overlapping calibration grids simultaneously. This is then improved using pose estimation and bundle adjustment-type algorithms. The approach has been applied on a wide variety of central and non-central cameras including fisheye lens, catadioptric cameras with spherical and hyperbolic mirrors, and multi-camera setups. We also consider the question if non central models are more appropriate for certain cameras than central models.
The work was mostly carried out during my stay at the MOVI project in INRIA Rhone Alpes, France. |
|
|
Generic Cross-Camera
Structure-from-Motion ( Omnivis’04 )
We introduce a generic structure-from-motion approach based on a previously introduced, highly general imaging model, where cameras are modeled as possibly unconstrained sets of projection rays. We introduce a structure-from-motion approach for this general imaging model that allows reconstructing scenes from calibrated images, possibly taken by cameras of different types (cross-camera scenarios). |
|
|
3D
Reconstruction from Multi-Scale Images
Adaptive fusion of new information in a 3D urban scene is an important goal to achieve in computer vision, graphics, and visualization. In this work we acquire new image pairs of a scene from closer distances and extract 3D models of successively higher resolutions. We present a new hierarchical approach to register these texture-mapped 3D models with a coarse 3D texture mapped model of an urban scene. This is done by matching images at different scales using an algorithm similar to SIFT. A mesh-merging algorithm is also presented to perform a smooth merging while registration. We present the results of our hierarchical algorithm for adaptive enhancement of a mural inside the UCSC Campus by registering data that differ in scale by a ratio of 1:100.
|
|
|
Simplification of LiDAR data (Report)
We applied Real Time Optimally Meshing algorithm (ROAM)
algorithm on the LiDAR data of UC,
|
|
|
Semi-Automatic 3D Reconstruction of UCSC (Report)
We model and reconstruct the buildings and terrain in UCSC campus using a semi-automated algorithm. A coarse texture mapped 3D model for the campus is first constructed using various sensors. For example, we model the terrain, height of the buildings and wall textures using the information obtained from DEM, LiDAR and cameras respectively.
|
|
|
Multiview motion analysis and bundle
adjustment ( From multiple images & pair of images )
In my computer vision and image processing courses I was introduced to the problem of 3D reconstruction from images. As part of course projects I implemented 3D reconstruction algorithms using multiple images along with bundle adjustment routines to refine the results.
|
|
|
Visualization of Geospatical Data and VGIS ( IEEE Vis’02 Demo 1, Demo 2, SPIE’02, CGI’02,
In the first two years of my masters program I worked on the Next generation, 4D distributed modeling and visualization project. |
