Predicting Artist Drawing Activity via Multi-camera Inputs for Co-creative Drawing

This paper presents the results of computer vision experiments in perception an artist drawing with analog media (pen and paper), aim to contribute towards a human-robot co-creative system. Using data gathered from user studies artists illustrators, two types CNN models were designed evaluated. Both use multi-camera images surface as input. One predicts artist’s activity (e.g. are they or not?). The other model position pen on canvas. Results different combination input sources presented. overall mean accuracy is 95% (std: 7%) for predicting when present 68% 15%) drawing. pen’s canvas squared error (in normalised units) 0.0034 0.0099). These development autonomous robotic system which aware at work via camera based In addition, this benefits more fluid physical digital workflow creative content creation.