Visually-based Prediction of Artist’s Drawing

Abstract:

This paper describes recent work in the development
of a co-creative human-robot drawing system, which observes an
artist’s drawing process in real-time. Using the data gathered in
a recent pilot study, a series of models were trained in order
to recover the current state of the artist’s drawing behaviour
and pen attributes from a multi-camera multi-perspective setup,
aligned to a “ground truth” dataset obtained from a drawing
tablet. Experiments, carried out with two computer vision models
based on a CNN architecture, form a baseline for future, more
sophisticated models.

Download PDF