A Hybrid Approach for Eye-centre Localization for Estimation of Eye-Gazes using Low-cost Web Cam

Gaze estimation is one of the recent popular research topics in the domain of computer science/engineering for vision-based human-computer-interaction. It has huge potential for application in assessment of physical condition of drivers while driving, controlling robots and prosthetics, surveillance etc. For successful gaze estimation the eyes have to be detected perfectly and this will lead to efficient eye centre localization. Several approaches have been used by researchers for eye and eye centre detection. The present work uses Haar-like cascade classifier for effective face and eye detection. This method has the advantage of lower computational load, faster processing due to the associated AdaBoost algorithm. High success rate can be easily achieved in several environments using proper training sets. However, for eye centre localization an improved version of Hough transform in two dimensional parametric space has been used as it is very simple to use and implement practically. This hybrid approach has been successfully tested using low-cost webcams in different lighting conditions with and without spectacles.