Research Interests:

  Signal, image/video processing and communication, computer vision, medical imaging;

  Multi-dimensional scanning, 3D measurement and other multimedia applications;

  Camera calibration methods and Computational Geometry;

  Computer graphics/animation and image based rendering;

  Medical Imaging & Biomedical Instrument Development;

 

Past Research:

  Subpixel measurement of target displacement

  Optical Flow

  Stereo disparity

  Image filtering

 

Current Research:

3D Scanning of moving objects based on "one snap" FTP methods;

Fully Automated cameras calibrations methods;

  Chessboard grid calibration based on a double Radon Transform.

Colour Space Transformation;

 

  3D Scanning of moving objects based on "one snap" FTP methods

 

My Ph D project consists in designing a 3D scanner to perform surface measurement of moving objects. The 3D scanners are gaining huge spreading in the market today, but the great majority of them can only scan surfaces of stationary objects. The innovation of the project consists in extracting the surface skin of an object without imposing it to stop.
The calibration grid	The scanner
You will find here following some preliminary results:
The otput with the texture	Click on The image to enlarge it !   The Vrml file of the output is about 19MB even zipped it was to big to link here, If You need It please e-mail me

 Chessboard grid calibration based on a double Radon Transform

The goal of this work was:   Minimize the time needed to calibrate a 3D FTP Scanner   automatizing the Calibration Process   Create a robust algorithm   insensitive to the aberration caused by the lenses   insensitive to generic backgrounds
	When the scanner is calibrated? Camera & Projectors - Intrinsic parameters: o Focal length o Principal point o Skew coefficient (angle between the x and y pixel axes ) o Distortions (radial and tangential) - Extrinsic parameters: o Rotations o Translations
Innovation of the method Improvements   It can extract corners even if the chessboard is partially occluded or if the corners are outside the image   It can exclude all the corners which are not on the chessboard
The Double Radon Transform
The Klaus Kohlmann algorithm   It consists in a 2D Hilbert transform   Limit of the Algorithm           applicable only in the range from - 5° to 5° rotations
Combining the two algorithms potentialities
In this case DRT is fundamental to give a good estimate of the reference frame connected to chessboard grid
Computing than a suitable prospective transform to obtain the edge of the grid be parallel to the axis image
Than Klaus Kohlmann algorithm will give us the right estimation of the corners positions with subpixel Accuracy
Example: Click on the image to open the movie