EyesOn Mobile Eye Tracking

3 Infrared Eye Tracker Hardware 4 Generation 1 4 Generation 2 4 Gaze Interpretation 5 Noise reduction 5 Corneal Reflection Detection 5 Pupil contour detection 6 Ellipse fitting 7 Calibration 7 Screen detection 7 Sources of Error 8 Parallax analysis 8 Device analysis 9 Discussion 10 3 EyesOn Mobile Eye Tracking Abstract Eye tracking is an extremely valuable resource for behavioral research and the next generation of human computer interaction, especially for handicapped individuals. However, obtaining robust, high quality eye-tracking data can be enormously expensive, and low cost alternatives can be inaccurate and unable to be used as computer input devices. High end mobile eye tracking systems, such as those manufactured by Tobii®, can cost as much as twenty thousand dollars or more. I hope to deliver a method for creating a very low cost, easy to assemble mobile eye tracking unit which uses a USB interface, and to create robust software to analyze the video data streams. The novel contribution is a method for screen detection on the mobile eye-tracker by the use of a hot swappable filter and Infrared LEDs, and rigorous analysis of the accuracy of the tracker in various lighting conditions. This will allow real time eye-tracking as a method of computer input on a head mounted mobile eye-tracker, and create the opportunity for quality research at a low, affordable cost.Eye tracking is an extremely valuable resource for behavioral research and the next generation of human computer interaction, especially for handicapped individuals. However, obtaining robust, high quality eye-tracking data can be enormously expensive, and low cost alternatives can be inaccurate and unable to be used as computer input devices. High end mobile eye tracking systems, such as those manufactured by Tobii®, can cost as much as twenty thousand dollars or more. I hope to deliver a method for creating a very low cost, easy to assemble mobile eye tracking unit which uses a USB interface, and to create robust software to analyze the video data streams. The novel contribution is a method for screen detection on the mobile eye-tracker by the use of a hot swappable filter and Infrared LEDs, and rigorous analysis of the accuracy of the tracker in various lighting conditions. This will allow real time eye-tracking as a method of computer input on a head mounted mobile eye-tracker, and create the opportunity for quality research at a low, affordable cost. Infrared Eye Tracker Hardware In this section, the design of the EyesOn eye-tracking hardware is described in a way that shows the evolution of the system to its final form. This approach provides insight into principles, decisions, benefits, and limitations of the system. The first design consideration after having chosen to use a head-mounted system was the configuration of the head gear. The most significant issue was where to mount the cameras. Two Logitech Pro 9000® cameras were utilized as the video devices. The USB cameras were stripped of all plastic and nonessential components so ensure they were as light as possible. An infrared LED was affixed to the USB camera board with the help of a resistor, which reduces the current through the LED to ensure the light intensity projected by the LED is not harmful. The infrared blocking filter in the USB camera was removed from the eye camera, and the filter was replaced with an infrared passing filter. The lens of the scene camera was placed as close to the physical eye as possible without occluding vision, in order to minimize the induced parallax error. The smallest distance achievable from the affixed camera to the eye was approximately 1.2 inches (3 cm). Some commercial units utilize hot mirrors, bits of plastic which reflect infrared light but are completely passing to other wavelengths, in conjunction with their eye-trackers, to reduce the error induced by camera sway. However, these products can be expensive and difficult to obtain in small quantities. In order to reduce costs, the USB camera was affixed using steel reinforcement wire to point with a direct line of sight towards the eye. Since the camera is extremely light, and the steel reinforcement cable is comparatively strong and rigid, there is 4 EyesOn Mobile Eye Tracking not a significant amount of camera sway due to head movement, and what sway there is quickly tapers off. The primary disadvantage of a boom arm design is that a portion of the visual field is blocked by the camera and the armature. Because there is only a small extent of visual occlusion and peripheral positioning of the camera, this is an acceptable compromise. In fact, because these components are attached to the head gear and thus static in the user's visual field, they are easily ignored just as the frames of normal eye glasses are ignored. (a) (b) Figure 1: Eye tracker and the captured images. (a) Head-mounted eye tracker. (b) Image of the user’s right eye illuminated with infrared light. Note the clearly contrasted dark pupil and the best fitting ellipse corresponding to the algorithms analysis of the image. The second generation prototype was very similar. A rubber goggled strap was used to snugly hold the device to the subjects head, the hot-mirror was removed from both the scene and the eye camera, and a hot-swappable filter was placed on the scene camera, to allow the camera to quickly switch between the visible and infrared light spectrums. A detailed step by step guide on the construction of the device will be covered in appendix B, and a detailed step by step guide on the use of the device will be covered in appendix C. 5 EyesOn Mobile Eye Tracking Gaze Interpretation Presented in this section is a robust eye-tracking algorithm, modified from the Starburst algorithm, which combines feature-based and model-based approaches to achieve a good trade-off between run-time performance and accuracy for dark-pupil infrared imagery. The goal of the algorithm is to extract the location of the pupil center and the corneal reflection so as to relate the vector difference between these measures to coordinates in the scene image. The algorithm does the following: • Locate and remove the corneal reflection from the image using thresholding. • Detect pupil edge points using an iterative feature-based technique. • Remove bad pupil edges using the Random Sample Consensus (RANSAC) algorithm. • Find the best fitting ellipse to the remaining feature points, the center of the ellipse is the pupil center.