Gaze Estimation Using Human Joint Rotation Angel

Currently, eye trackers can estimate with high precision using eye images. This method requires high-quality eye images, and thus cannot estimate using low-quality eye images, which have low resolution, illumination changes, and occlusions. Other methods use pose information, including head direction, body direction, and head motion. This method obtains pose information using the Eigenspace method, and thus requires a calibration phase. In addition, the eye direction using this method is only horizontal. We propose a gaze estimation algorithm that does not require high quality eye images. The purpose is estimation without using eye images, with an error of less than ten degrees. Experimental results show that the method can estimate gaze direction without using eye images.