ENERGY EFFICIENT GAZE TRACKING METHOD AND APPARATUS BASED ON SIMPLIFICATION OF CASCADE REGRESSION FOREST

摘要

The present invention relates to an energy-efficient pupil tracking method based on the simplification of a cascade regression forest, and more specifically, to a pupil tracking method. generating an initial CCD-RF (a cascade coarse-to-fine regression forest) learned for a pupil position in a to-fine manner; (2) generating a lightweight CCD-RF by applying rule distillation according to the rule contribution based on feature importance to the generated initial CCD-RF; and (3) predicting a pupil position from an image using the lightweight CCD-RF. In addition, the present invention relates to an energy-efficient eye tracking device based on the simplification of a cascade regression forest, and more specifically, as an eye tracking device, it is composed of a cascade multi-layer using a plurality of regression forests, and coarse-to a learning unit that generates an initial CCD-RF (a cascade coarse-to-fine regression forest) learned for the pupil position in a fine manner; a weight reduction unit for generating a lightweight CCD-RF by applying rule distillation according to a rule contribution based on feature importance to the generated initial CCD-RF; and a prediction unit for predicting a pupil position from an image using the lightweight CCD-RF. According to the energy-efficient eye tracking method and apparatus based on the simplification of the cascade regression forest proposed in the present invention, it is composed of a multi-layer cascade method using a plurality of regression forests, Lightweight CCD-RF is generated by applying rule distillation according to rule contribution based on feature importance to the learned initial CCD-RF (a cascade coarse-to-fine regression forest). By predicting the pupil position from the image using