Abstract
Eye tracking is a well-established tool that is often utilised in research. There are currently many different types of eye trackers available, but they are either expensive, or provide a relatively low sampling frequency. The eye tracker presented in this paper was developed in an effort to address the lack of low-cost high-speed eye trackers. It utilises the Graphical Processing Unit (GPU) in an attempt to parallelise aspects of the process to localize feature points in eye images to attain higher sampling frequencies. Moreover, the proposed implementation allows for the system to be used on a variety of different GPUs. The developed solution is capable of sampling at frequencies of 200 Hz and higher, while allowing for head movements within an area of 10×6×10 cm and an average accuracy of one degree of visual angle. The entire system can be built for less than 700 euros, and will run on a mid-range laptop.
Highlights
Eye trackers are useful for a number of research applications, such as research in reading and dyslexia, language processing and mental health disorders (Duchowski, 2002; SensoMotoric Instruments, n.d.)
The effect of head position on the precision and accuracy was investigated at a sampling frequency of 100 Hz while the position of the eye tracker was changed in relation to the participant to simulate head movements
Eye tracking systems contain noise and the purpose of the test with artificial eyes was to determine the precision that is obtainable if the continuous head movements and eye tremors that are present in human participants were completely eliminated, thereby determining the amount of noise produced by the eye tracking system
Summary
Eye trackers are useful for a number of research applications, such as research in reading and dyslexia, language processing and mental health disorders (Duchowski, 2002; SensoMotoric Instruments, n.d.). When one considers that an eye video consists of many thousands (even millions) of pixels, it stands to reason that a dedicated device, such as the GPU, could be highly advantageous in parallelising the image processing tasks in eye tracking. This points to the SIMD architecture of the modern GPU, which allows operations to be performed on many pixels simultaneously. As mentioned previously, Mulligan (2012) proposed a solution making use of CUDA that is capable of achieving 250 Hz on a 640×480 image, with average accuracies close to 0.5 degrees of visual angle In another example, Duchowski et al (2012) made use of the GPU to build real time heat maps. They found that pupil tracking can be efficiently performed at framerates up to 530 Hz using a state-of-the-art GP-GPU
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
