Gaze estimation has received a great deal of attentions, due to its wide range of applications including studies of psychology, market and advertising analysis, medical researches, human-computer interactions, etc. It has become a hot topic in pattern recognition and human-computer interaction. With some published papers and commercial systems, gaze estimation under well controlled environment is relatively mature, providing high accuracy and meeting some kinds of practical applications. However, existing gaze estimation techniques still have many limitations, such as complicated calibration process, restricted environment and head movements. In this thesis, the issues of gaze estimation based on image analysis under several kinds of non-invasive devices are studied. This work focuses on the main difficulties in gaze estimation research and attempts to provide practical solutions. The main contributions of this thesis include the following issues: 1. Gaze estimation under condition of single camera and single light source. To solve the problem that the calibration procedure in most current gaze estimation methods are tedious when only single camera and single light source are used, we propose a novel gaze estimation method with one-point calibration. In our approach, statistical models of multiple points on the screen are built in advance and interpolation-based method is used to estimate the PoR(Point of regard) of the user on the screen. The main contributions of this paper are: 1) a coarse-to-fine algorithm for pupil location is presented, which improves the adaptability to eye glasses, eyelashes, occlusion and image blurring; 2) a novel one-point calibration gaze estimation model based on statistical method is proposed, which reduces the complexity of the calibration procedure; 3) incremental learning method is used to update the model, which could improve the adaptability of different users and head movements. The proposed method is effective for different users with different head movements. 2. Gaze estimation under condition of single camera and two light sources. Existing appearance-based and feature-based methods both have achieved impressive progress in the past several years, while their improvements are still limited by feature representation. Therefore, we propose a novel descriptor combining eye appearance and pupil center-cornea reflections (PCCR). The hybrid gaze descriptor represents eye structure from both feature level and topolog...
修改评论