Image Processing for Digital Twin Technology
Abstract
Consider the problem of constructing a digital twin by taking a two dimensional CAD drawing such as a building floor plan, and and augmenting it with new three dimensional data. The new three-dimensional data is a point cloud in some unknown coordinate system. It can come from Laser range finding or from comparing common features in multiple images from a video stream, but it might also be useful to match Lidar-based 2D data against a CAD drawing.
An algorithm that has worked well on real world data is first to analyze the point cloud to decide which way is up, then apply polyline-based matching. The "which way is up ?" algorithm involves optimization of a "how well clustered" metric. The polyline-based matching is applied after projecting the point set onto a floor plane, and analyzing vertical histograms to decide which parts of the projected image most resemble walls. The vertical histograms are also used to correct for residual tilt due to inaccuracy in the "which way is up ?" result.
Short Biography
John Hobby earned his PhD in computer science from Stanford
University in 1985 under Donald Knuth. Since then, he has
been at Bell Labs in Murray Hill, New Jersey. It was originally
known as AT&T Bell Labs and is now Nokia Bell Labs.
John's work covers a wide variety of practical algorithms,
and he is now in the Artificial Intelligence Research Lab.
