Learning gaze biases with head motion for head pose-free gaze estimation

作者：

Highlights：

• Appearance-based gaze estimation with head motion is decomposed into subproblems.

• Subproblems are solved by compensating for two types of estimation biases.

• The compensation method only requires a 5-second video for training.

• Eye images for different head poses are aligned via rectification and optimization.

• We achieve a gaze estimation accuracy of 3° with free head motion.

摘要

•Appearance-based gaze estimation with head motion is decomposed into subproblems.•Subproblems are solved by compensating for two types of estimation biases.•The compensation method only requires a 5-second video for training.•Eye images for different head poses are aligned via rectification and optimization.•We achieve a gaze estimation accuracy of 3° with free head motion.

论文关键词：Gaze estimation,Free head motion,Head pose compensation,Appearance-based approach

论文评审过程：Received 14 June 2013, Revised 18 December 2013, Accepted 9 January 2014, Available online 19 January 2014.

论文官网地址：https://doi.org/10.1016/j.imavis.2014.01.005