CONTROLLING COMPUTER USING EYE WITH HIGH ACCURACY AND FAST PERFORMANCE WITH A STANDARD WEBCAM WITHOUT SENSORS OR INFRARED LIGHTS

Abstract

Through human computer interaction, Machines can help and assist humans with their daily activities in a smoother way; this includes home use, office use for any age level. This project’s purpose was to determine an optimized and simplified way to extract eye pupil position in a
video frame (Simplified compared to current complex and expensive algorithms) for enhancing human interaction systems by adding vision features. Last year, the Real time auto-focus algorithm has achieved an accuracy of 50% - 75% with high performance cost. Other newly
researched algorithms proved close robustness and accuracy with infrared lights and expensive sensors causing a CPU load of 60 -80% (2GHZ Core Duo with 1GB of RAM), a RAM usage of 700MB with 20% tolerance and 700$ for the sensors and IR equipments. This year, the project’s
purpose was to enhance the performance by replacing the complex hardware equipments and software algorithms with a simple web cam and a set of Fourier filtration steps and AI procedures. A performance evaluation was performed to test the new approach, using a standard
webcam of 1.3 megapixels and a 2.2 GHZ CPU core 2 Duo and 1 GB of RAM. The light conditions were changed to be normal, extra light, dim light with a noise –Multiple heads in the background. The results were compared with the latest algorithms that were found during the last
year. The CPU load didn’t pass 25%, the RAM was used by 300 MB rate and the cost was 5$ - the webcam cost, with high accuracy and tolerance up to 3%. All resources were used in smaller ratio with height accuracy.
ratio with height accuracy.