A glasses tracking interface for human-robot interaction

This paper introduces a glasses tracking interface as an alternative to tracking the human head, which can be used in human-robot interaction. The main idea of this study is originated from the fact that the face of a glasses frame resembles two coplanar conics, which can be extracted and approximated simply through visual processing. We mainly address, in this paper, how the 3-D pose of two-conics object can be retrieved from a correspondence set of two conics, which lie on an image plane and a model plane, respectively. To obtain the best solution of 3-D pose parameters especially in practical applications, an iterative optimization procedure is described to find optimum values for rotation parameters. Based on the solution, we also propose a glasses tracking interface composed of a camera with an infrared-pass-filter and a glasses frame with reflective marker patterns. The proposed method is tested and discussed through some experiments for evaluating estimation accuracy.