Freehand Acquisition of Objects

K. Häming, G. Peters

subm.: J. Communication and Computer
INTERACT 2009
WIO 2006

Abstract

In many applications 3d models of real-world objects are required. We introduce a tool which allows an untrained user to take three images of an object freehand with a simple consumer camera. From these images a 3d model of the visible parts of the object is reconstructed. The problem we address is the reconstruction of an object especially from a short sequence of images with an uncalibrated camera which functions in a fast and reliable way. From a research point of view we propose solutions for three weaknesses of the state-of-the-art reconstruction pipeline: an improved SIFT-based feature detection, a two-stage RANSAC process facilitating a faster selection of relevant object points, and a novel texture mapping. From a practice point of view we present a user-friendly tool for fast acquisition and realistic visualization of 3d models of real-world objects. As it requires low-level technical equipment only and no technical knowledge of the user this tool may be attractive to a large number of people in the future.

WIO 2006 paper (pdf)

INTERACT 2009 paper (pdf)

Broadcast on WDR: Lokalzeit aus Dortmund, 29.10.2008

AVI (232 MB) A. Tchouev, G. Peters, K. Häming, G. Stamm, T. Pucyk

Research

Commonly the reconstruction pipeline looks like this. In the first step (image processing step) feature points on the object are detected independently in each input image. These features are matched between pairs of images. In a second step (reconstruction step) only those matches are filtered which are consistent with all images. From these remaining object points a 3d point cloud can be reconstructed. In a third step (texture mapping step) texture of the object surface is mapped onto the 3d point cloud resulting in the final 3d model. Our contributions to this reconstruction pipleline are marked in red: an improved SIFT-based feature detection, a two-stage RANSAC process facilitating a faster selection of relevant object points, and a novel texture mapping. [2009]

Some Results: 3D Models Reconstructed From Three Images Taken Freehand