Non-invasive gaze direction estimation from head orientation for human-machine interaction

Document Type

Conference Proceeding

Publication Date

6-1-2018

Department

Department of Mechanical Engineering-Engineering Mechanics

Abstract

Gaze direction is one of the most important interaction cues that is widely used in human-machine interactions. In scenarios where participants’ head movement is involved and/or participants are sensitive to body-attached sensors, traditional gaze tracking methods, such as using commercial eye trackers are not appropriate. This is because the participants need to hold head pose during tracking or wear invasive sensors that are distractive and uncomfortable. Thus, head orientation has been used to approximate gaze directions in these cases. However, the difference between head orientation and gaze direction has not been thoroughly and numerically evaluated, and thus how to derive gaze direction accurately from head orientation is still an open question. In this article, we have two contributions in solving these problems. First, we evaluated the difference between people’s frontal head orientation and their gaze direction when looking at an object in different directions. Second, we developed functions that can map people’s gaze direction using their frontal head orientation. The accuracy of the proposed gaze tracking method is around 7°, and the method can be easily embedded on top of any existing remote head orientation method to perform non-invasive gaze direction estimation.

Publication Title

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

ISBN

978-3-319-91250-9

Share

COinS