Fast and robust: 3D reconstruction of the environment from images

In close-range photogrammetry untrained users typically do not capture images in a pre-planned pattern, and even experts often need significantly more time for data acquisition when strict recording protocols must be followed. For some years, following the development of sensors, images can be accessed in a much easier way, e.g. when taken with a smart phone or downloaded from the Internet. The challenge with such images is that it is not known where they were taken from and which images actually overlap, both of which is a pre-requisite for 3D reconstruction.

In this project, we first automatically determine the image overlap based on an overlap graph and fast search algorithm running on so called a k-d- forest. Then, we solve the exterior image orientation parameters, referred to as structure from motion (SfM) problem in computer vision. While typical methods work in an incremental way, starting with two images and then adding the rest to the block sequentially, we suggest a global method, which does not need any initial values for the unknowns and is much faster. As it is known that global methods are more sensitive to blunders, we take care to find these gross errors by introducing special checks at all stages of the processing chain. In this way we are able to eliminate stereo pairs with a very short baseline, which yield imprecise results in depth, and can deal with images showing repetitive structure.