A new image-based method developed by computer scientists at UCL, Adobe Research and the University of Hong Kong captures the complexities of thin, bendy objects.The novel computational method can reconstruct wiry objects digitally from a few input images, and could be used by animators to depict wire- or cable-like objects, or by medical practitioners to examine networks such as blood vessels. “Image-based Reconstruction of Wire Art ” was presented at the annual SIGGRAPH conference by Lingjie Liu and Professor Niloy Mitra of UCL Computer Science, and collaborators.
Digitizing objects made from bent wires, such as jewellery and sculpture, remains a challenging problem because they consist entirely of one-dimensional elements. Wires are difficult to translate into a 3D model from a 1D image because of the lack of features, thin cross-section, and the fact they often cross on top of themselves.
Professor Niloy Mitra, coauthor of the research and professor of geometry processing at UCL, explains: “The problem with existing methods is that they return isolated sets of points locating the wiry object, with no meaningful information about how they are connected. We observe that knowing how an object is made, helps to reconstruct it. So in computing wiry objects we aim to directly recover wires, rather than its isolated points.”
The team were inspired by observing piles of tangled strings – climbing rope, to be exact. Their method first looks at the connectivity of the systems from multiple angles. Then in order to recreate the physical object digitally, they modeled 3D curve parts in sections, choosing short elements that could be assembled together into their full configuration. They exploit unique characteristics of wiry objects, such as their simplicity – they consist of only a few wires – and their smoothness – wires do not make sudden kinks -to digitally reconstruct the entire 3D wire composition.
In the study, the researchers demonstrated their new method on objects with varying complexity, each composed of one to three wires; the object examples included wire sculptures of elephants, birds and flowers. Compared to existing techniques, their method creates much higher resolution reconstructions which more accurately capture both the 3D geometry and the topology of the wires.
This could be applied to a number of fields which address thin structures, such as finding the connectivity of nerves in a neural network from a series of images. It could be particularly relevant for medical contexts where reconstruction of thin structures from a few views is required.
The paper, a video illustration and example code are available here: