Eye Tracking for HCI (via ordinary web cameras)

Hits: 0
Research Area: Uncategorized
Status: Not started  

The term “high-level languages” is due to the command syntax proximity to the human language. 
However, since the invention of computer mouse (interface for GUI), interaction between human and computer (Human Computer Interaction, HCI) knows no greater jump. Still, many attempts to improve HCI were made through many alternatives techniques: voice commands, computer mouse variations (touch screens, trackballs, pen pads, etc.), brain signals interpretation and eye tracking.

Touch screen:

Touch screens still require hand movement. Solutions as voice commands interpretation are inappropriate to the goal proposed. These systems input data are words and/or other sounds. A simple mouse movement is quicker and most accurate. Besides, the input data is prone to ambient interferences, pitch and diction, etc. Efficient brain signals classifier/interpreter needs expensive and/or non-ergonomic apparel attached to the user’s head:


So why do we keep on searching for another solution instead of just using what we already have: traditional mouse? Because eyes are faster…
Despite some Eye Tracking Systems needing equipment attached to the user’s head, like a cap or a pair of glasses with one (or a pair) of cameras, our effort converges to the development of an alternate (faster) interaction using regular web cameras, already embedded in screens.
The areas of application off such techniques range from regular HCI, to facilitate computer access to disabled and impaired people, analysis of attention (driver drowsiness, web browsing aim areas statistics, etc.), advanced interaction experiences with other similar systems and industrial machinery/computer systems interaction.


Most of Eye Tracking systems have three main phases: face localization, eye localization and eye tracking (position analysis). The first might be suppressed with more advanced algorithms.

Eyes location results:

Tracking measurement:

Main difficulties on getting the job (well) done: obviously, the human morphological diversities (eyebrows proximity to the eye, for example), eye glasses, accessories (hat cap), shadows and ambience illumination conditions.

Detection for different conditions and subjects: 

Unsuccessful detection:

The most relevant techniques for each phase of Eye Tracking
i) Face location: distribution of skin color model classification (chromatic color space); face shape template (geometric shapes); face features classification (of elements: eyes, nose, lips, etc.).
ii) Eyes location: template matching (geometric: eyes, eye lips), neural network classification, isocentric patterns, non-skin color classification (processed by previous distribution of skin color model.