Gaze point estimation on curved display by using session level calibration for flat screen displays
Most of the existing gaze tracking schemes with high accuracy and high speed depend on infra-red (IR) lights and multi-cameras, which leads to high complexity of apparatus and high cost. Besides, many proposed approaches hardly offer a full discussion and solution of eye blink issue. In this paper, we propose a novel gaze tracking scheme which is capable of tracking eye movements in high accuracy. Our scheme incorporates the eye corner information extracted using a novel eye corner detector. This detector is developed based on the Gabor Wavelet Transform and the Structure Tensor. Gabor Wavelet Transform decomposes an image in multi-scales and multi-orientations, thus is robust against lighting variation and tiny shift. We abstract the distribution statistics of the feature points in the eye region and re-express it as a connectivity graph. Based on such abstraction we propose a novel solution to the eye blink issue which obtains a high successful detection rate. After implementation, our scheme is proven to be accurate compared with the state of the art. Notably, only one web camera is employed in our scheme without any auxiliary light source or cameras.