RESTORE: REconStruction with Texture-based Object REfinement

The 3D reconstruction of real objects from multiple images is a recent and exciting research topic in the field of computer vision. At the same time the demand for 3D models in the movie and game industry is increasing.

A simple and cost-efficient way of reconstruction real objects is using only a single camera. In this project a hard- and software system is developed which makes the 3D reconstruction of real objects easy. The hardware consists of a turntable on which an object can be placed, a motor which drives the rotary table, a controller to control the motor as well as a calibrated camera for taking pictures of the object.

From the camera images a visual hull is first computed, which gives a first rough estimation of the 3D reconstruction. This model is then refined using only the texture of the object.

With the silhouettes of an object from multiple camera views, the first step of the 3D reconstruction is computing the visual hull. That is the maximum volume in 3D space which may have produced them. This step requires the camera projection matrix from every camera image to be known.

RESTORE is released under the terms of the MIT License.