Abstract
As the demand for human-friendly computing increases, research on pupil tracking to facilitate human–machine interactions (HCIs) is being actively conducted. Several successful pupil tracking approaches have been developed using images and a deep neural network (DNN). However, common DNN-based methods not only require tremendous computing power and energy consumption for learning and prediction; they also have a demerit in that an interpretation is impossible because a black-box model with an unknown prediction process is applied. In this study, we propose a lightweight pupil tracking algorithm for on-device machine learning (ML) using a fast and accurate cascade deep regression forest (RF) instead of a DNN. Pupil estimation is applied in a coarse-to-fine manner in a layer-by-layer RF structure, and each RF is simplified using the proposed rule distillation algorithm for removing unimportant rules constituting the RF. The goal of the proposed algorithm is to produce a more transparent and adoptable model for application to on-device ML systems, while maintaining a precise pupil tracking performance. Our proposed method experimentally achieves an outstanding speed, a reduction in the number of parameters, and a better pupil tracking performance compared to several other state-of-the-art methods using only a CPU.
Highlights
Human gaze perception during a social interaction is an important non-verbal factor facilitating communication between people
In the field of machine learning (ML) in particular, human gaze perception is essential as an interface for a human–machine or human–computer interaction (HCI)
Cascade pupil estimation technique consisting of a fast and accurate cascade deep regression forest (RF) instead of a deep neural network (DNN)-based pupil estimation; estimation of the pupil location per layer in a coarse-to-fine manner and a refinement of the pupil area from the estimated location and eye area of the previous layer; RF simplification removes low-importance rules according to the feature importance during the process of reconstructing a rule set; the rule distillation proves that the performance is not degraded even if numerous rules and parameters are eliminated; pupil consistency checking robust to fast movements of the face and maintain an accurate pupil tracking;
Summary
Human gaze perception during a social interaction is an important non-verbal factor facilitating communication between people. The pupil tracking approach designed through these motivations can be applied to low-power devices because it is interpretable by reducing the rules of CCD-RF, and energy-efficient due to reduced computation. The of this study is to design lightweight pupil tracking model that is interpretable and IMLgoal explores and investigates howaML models are created or complemented, allowing humans energy efficient for application to an on-device. We divide the related studies into the to access and interpret the results of internal logic and algorithms, allowing humans to understand research trend of IML and pupil-tracking based on a light. IML can be divided into a transparent model revealing how it functions, and a post-hoc explanation that describes why a black-box model behaves depending on the model’s
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
