Research On Gaze Estimation For Human-Computer Interaction

HCI is an important component of computer system. Ever since the emergence of computer, HCI has gone through three stages of development:command interaction stage, GUI stage and natural HCI stage. Natural HCI is the HCI with intuitive interaction manner, which can be performed with only everyday skills. Gaze estimation based HCI is one kind of natural HCI. With a promising application future, it has become a popular research topic.Gaze-Estimation-based HCI (GEHCI) is one of the next generation of HCI, in which user’s gaze is estimated for interaction. Combined with GUI, GEHCI is more natural and faster than the interaction with mouse, furthermore, GEHCI can be performed totally without hand. As a consequence, GEHCI is supposed to have a wide range of potential applications. Currently, a number of products for GEHCI is available on the market, but their prices are too expensive to make them civilian products. The costs of these products are mostly associated with hardware, because the gaze estimation methods rely on high-quality digital cameras. Therefore, the research on cheap-equipment-based gaze estimation is of great significance.To promote the popularization of GEHCI, we study a cheap-equipment-based gaze estimation technology for HCI in this thesis. The main research works and contributions are summarized as follows:I. We proposed a Mapping and Iris-center-correlation-based Gaze Estimation algorithm (MIGE). The algorithm first calculates the mapping from iris center in eye image to gaze position in scene image. Then this mapping is used to estimate gaze positions. Locating the iris center in eye image is the key process in the algorithm. The process is as follows:the region grow method with fixed region shape and position adjustment is performed on the eye image to detect the iris region, and then an ellipse is fitted from the boundary of the iris region, after that the iris center is located by modifying the ellipse center with a correction offset, which is calculated from the eye geometry model proposed in this thesis. Through experiments, we show that MIGE algorithm outperform the similar methods.2. We proposed a Local-Extremum-based Blink Detection algorithm (LEBD) and a Patterned-Blinks Classification algorithm (PBC), and we implement three interaction movements based on the two algorithms. To make interaction more natural and more efficient, in this work, eye blink is combined with gaze to implement interaction movement. To detect a user’s blinks, we propose a robust, accurate and real-time algorithm called Local-Extrmum-based Blink Detection. We further propose a patterned-blink classification algorithm based on LEBD. Experiment results show that, after short-course training, uses can master the two patterned-blinks, i.e. double blink and long open, and the algorithm can get high detection accuracy. With the above two algorithms, the three interaction movements we implemented are more efficient than existing ones.3. A prototype system of GEHCI is developed. To verify the effectiveness of MIGE, LEBD and PBC algorithms, we develop a prototype system of GEHCI. The system can be used only with cheap equipment and visible light environment, and provides a friendly user experience.