The goal of this work was to develop a complete automatic pipeline to capture, process and render point-based models from images of real objects.

Duration:

09/2003 - 11/2003.

Personal Role:

Developing the complete image based modeling pipeline from image acquisition and calibration to reconstruction and point color estimation. Also developing the hardware accelerated rendering pipeline.

Details

In this work we developed a complete automatic pipeline to capture, process and render point-based models from real objects. The input to our system is a videosequence showing different perspectives of the object to model using a variety of imaging devices ranging from web-cams to high-end digital camcorders and computer vision cameras.

The proposed solution starts with a calibration of the input sequence based on feature detection and a divide and conquer linear calibration technique. The calibration information is then used to obtain a cloud of points of the surface of the object using a method inspired by the visual hull and voxel carving techniques. Then a post-process of the points is performed to smooth the reconstructed points positions and normal orientations.

Finally a hardware accelerated multi-resolution point-based rendering pipeline is used to obtain high quality images of the reconstructed objects at interactive frame rates.