A New Approach to an Old Problem: The Reconstruction of a Go Game through a Series of Photographs
This work addresses the domain-specific problem of automating Go game analysis from visual data, representing an incremental improvement in computer vision for board games.
The authors tackled the problem of automatically reconstructing a Go game's move sequence from a series of photographs by developing techniques for grid line detection, movement tracking, viewpoint approximation, and stone detection, resulting in a functional algorithm for full sequence reconstruction.
Given a series of photographs taken during a Go game, we describe the techniques we successfully employ for pinpointing the grid lines of the Go board and for tracking their small movements between consecutive photographs; then we discuss how to approximate the location and orientation of the observer's point of view, in order to compensate for projection effects. Finally we describe the different criteria that jointly form the algorithm for stones' detection, thus enabling us to automatically reconstruct the whole move sequence.