Living with paralysis is not a rare phenomenon. About 15% of the world's population lives with some form of disability. Of those, 2-4% experience significant functional impediments. Their daily activities are highly dependent on their caregivers. Thanks to advances in electric and mechanical devices, the disabled now have tools to become more independent. Among different control signals, gaze is becoming increasingly popular due to its high efficiency, ease of use, applicability to different types of activities, and robustness.

In this thesis, we describe systems that improve gaze-based interaction in both screen-based and 3D environments. First, we describe an improved method for dwell-based gaze typing using the user's past input history. We propose a probabilistic generative...[ Read more ]

Living with paralysis is not a rare phenomenon. About 15% of the world's population lives with some form of disability. Of those, 2-4% experience significant functional impediments. Their daily activities are highly dependent on their caregivers. Thanks to advances in electric and mechanical devices, the disabled now have tools to become more independent. Among different control signals, gaze is becoming increasingly popular due to its high efficiency, ease of use, applicability to different types of activities, and robustness.

In this thesis, we describe systems that improve gaze-based interaction in both screen-based and 3D environments. First, we describe an improved method for dwell-based gaze typing using the user's past input history. We propose a probabilistic generative model for gaze, which enables us to incorporate gaze behavior and past input history through Bayes' rule. We evaluate our model on both able-bodied subjects and subjects with a spinal cord injury. Compared to the standard method, we achieved 41.8% and 49.5% faster speed, in each population respectively.

Second, we describe a method to increase the robustness to head movement of remote eye tracker by incorporating the interaction history. We use estimated gaze targets inferred from the user's interaction to adjust the calibration parameters of the eye tracker online. We reduce errors by up to 43% as the head moves over a 20 cm range.

Third, we describe a method for feeding robot control that combines gaze and push button input. This combination not only speeds up the selection but, more importantly, reduces the number of movements required of the user.

Finally, to enable gaze-based interaction when people are mobile in the environment, we propose a 3D gaze estimation algorithm using a mobile eye tracker. This algorithm estimates 3D gaze in world coordinates by combining human pose estimated through SLAM, 2D gaze vectors estimated by a head mounted eye tracker, and the 3D environment context. We achieve a 2.8° accuracy over a 3:4 m × 4 m region.