A robust real time eye tracking and gaze estimation system using particle filters

Tariq Iqbal, University of Texas at El Paso


Eye tracking and gaze estimation techniques have been extensively investigated by researchers in the computer vision and the psychology community for the last few decades. Still it remains a challenging task due to the individuality of the eyes, variability in shape, scale, location, and lighting conditions. Eye tracking has many applications in neuroscience, psychology, and human-computer interaction. Gaze estimation plays a vital role in the field of human attention analysis, human factors in industrial engineering, marketing and advertising, human cognitive state analysis, gaze-based interactive user interfaces, and monitoring driver vigilance systems. Eye-tracking technology was originally a pioneer method in reading research, but has led to the development of many different methods to track eye movements and estimate gaze. Historical eye-tracking technologies include Electrical Oculography (EOG) and Coil Systems, which require additional electrical hardware mounted on the skin or specialized contact lenses to measure the eye movements. These methods proved quite invasive, instead current commercial eye trackers and gaze-estimation systems use video images of the eyes along with additional hardware, such as infrared sensors. In this thesis, we design and implement a low-cost eye-tracking system using only an off-the-shelf webcam. This eye-tracking system is robust to small head rotations, changes in lighting, and all but dramatic head movements. We also design and implement a gaze-estimation method which uses a simple yet powerful machine learning-based calibration technique to estimate gaze positions. The empirical evidence shows that our implemented eye tracker can track both eyes successfully and detect the pupil accurately in about 83% of frames regardless of the head movements and different lighting conditions. Additionally, we observe that the overall accuracy of the gaze estimation system is about 88% of the times throughout our experiments - albeit in extremely low resolution.

Subject Area

Computer science

Recommended Citation

Iqbal, Tariq, "A robust real time eye tracking and gaze estimation system using particle filters" (2012). ETD Collection for University of Texas, El Paso. AAI1518202.